thekb.eu — Economy & Market

GLM-5.2 leads open weights models and sits at #3 overall on GDPval-AA, a real-world agentic work benchmark

Mon, 22 Jun 2026 00:00:00 GMT

Artificial Analysis — an independent AI model evaluation platform — published (X/Twitter thread from June 22, 2026 + detailed model page) a comparison placing **GLM-5.2**, the latest model from **Z.ai** (Zhipu AI), at the top of **open weights** models and **#3 in the overall ranking** of **GDPval-AA**. This benchmark measures performance on **real-world, economically valuable knowledge work**, through **long-horizon and multi-turn** tasks, designed as genuine professional tests (for example, a retail store supervisor's daily task list, or an IEC technical document) covering both professional and creative work. GLM-5.2 scores **1524 Elo**, behind only **Claude Fable 5 (1783)** and **Claude Opus 4.8 (1615)**, and on par with **GPT-5.5 in xhigh setting (1509)**. More importantly, it dominates the open-weights field by a **wide margin**: the next-best open model, **MiniMax-M3**, scores only **1408**. GLM-5.2 also outperforms several proprietary models — **Gemini 3.5 Flash (1357)**, **Qwen 3.7 Max (1289)**, and **Muse Spark (1158)**. The **agentic** nature of the tasks is emphasized: GLM-5.2 averaged **~31 turns per task** across **1,999 matches**. Artificial Analysis's methodology consists of giving the **same briefs** to GLM-5.2 and three proprietary frontier models (Fable 5, GPT-5.5, Gemini 3.5 Flash), then **rendering each deliverable exactly as produced**. The result is consistent across the company's own indices: GLM-5.2 is **#1 among open weights** on the **Intelligence Index**, **#3 on the Agentic Index**, and **#3 on AA-Briefcase** (where it is the top open model, ahead of GPT-5.5 xhigh and behind only Fable 5). The model page rounds out the picture: GLM-5.2 is a **Mixture of Experts** with **753 billion parameters** (of which **40 billion active**), a **reasoning model** with **1M token** context, distributed under **MIT license** (commercial use allowed, weights on Hugging Face), released on **June 16, 2026**. On the economics side: **$1.40 / $4.40** per million tokens (input/output), a cache-hit rate of **$0.26** (-81%), a throughput of **106.3 tokens/s**, and a time to first token of **1.36s**. The message conveyed by the numbers is clear: that an **open weights** model at this price rivals the proprietary frontier on **genuinely useful agentic work** constitutes, according to Artificial Analysis, *"a real step for open models."* The open/proprietary convergence is no longer just about academic tests, but about the economic value produced under agentic conditions.

A frontier without an ecosystem is not stable

Fri, 12 Jun 2026 00:00:00 GMT

Satya Nadella, CEO of Microsoft, publishes on X a reflection on "the future of the firm" in an AI-driven economy. His starting thesis: this transition differs from any previous platform shift. Until now, digital systems augmented human capital; for the first time, it is possible to create a genuine **cognitive loop** between people and machines. What is at stake is not a tool, but the way organizations continue to learn, build their IP, differentiate themselves, and thrive in a world where AI models absorb and commoditize the expertise of individuals and organizations. Nadella proposes a central distinction: every company will need to build **human capital** (knowledge, judgment, relationships, ingenuity, pattern recognition) and **token capital** (the AI capability it builds and owns). Human capital does not lose value as token capital grows — on the contrary, it gains value: human agency is the engine driving the growth of token capital. Without human direction, "compute runs in circles." The real opportunity, then, is not choosing the best model, but building a **learning loop** on top of the models, where the two capitals compound. One can offload a task, or even a job, but never one's learning. This requires a new architecture in which every company builds agentic systems that improve over time while retaining control of its IP. The sovereignty test: being able to replace a "generalist" model without losing the expertise of the "company veteran." Three building blocks: **private evals** measuring improvement on the outcomes that matter to the business (not external benchmarks), **private RL environments** trained on real internal traces, and a **base de connaissances** making institutional memory queryable. This loop becomes the firm's new IP — a "hill climbing machine" that compounds: each improved workflow produces a better training signal, accelerating the accumulation of unique tacit knowledge, and creating an advantage that is difficult to replicate. Nadella concludes with a political-economy warning: a world in which a handful of models capture all the value will not be socially tolerated. He invokes the first wave of globalization, which hollowed out industrial economies through offshoring, as a cautionary tale. The priority must be to build a **frontier ecosystem**, not merely a **frontier model**, so that value diffuses to every company, sector, and country — the platform "ethos" he claims, and the only stable equilibrium worth building together.

Claude Fable 5 and Claude Mythos 5

Tue, 09 Jun 2026 00:00:00 GMT

On June 9, 2026, Anthropic announced the simultaneous launch of two models. **Claude Fable 5** is a "Mythos-class" model made safe for general use: its capabilities exceed those of any model Anthropic has publicly released, reaching state-of-the-art performance on nearly all benchmarks tested. **Claude Mythos 5** is the same underlying model, but with guardrails lifted in certain domains; it is restricted to a small group of cyberdefenders and infrastructure providers, deployed initially through Project Glasswing (in collaboration with the US government) as an upgrade to Claude Mythos Preview. Mythos 5 has the strongest cybersecurity capabilities of any model in the world. Both models are priced at $10 per million input tokens and $50 per million output tokens, less than half the price of Mythos Preview. To deploy quickly and safely, Fable 5 ships with deliberately conservative guardrails (classifiers): on certain topics, the query receives Opus 4.8's response instead. On average, they trigger in fewer than 5% of sessions. On capabilities, **software engineering**: Stripe reports that Fable 5 "compressed months of engineering into days," completing a 50-million-line Ruby codebase migration in one day (versus two months for a team). The model achieves the highest score among frontier models on FrontierCode (Cognition). **Knowledge work**: highest score of any model on Hebbia's Finance Benchmark (senior-level reasoning). **Vision**: state of the art; reconstructs a web app's source code from screenshots, completes Pokémon FireRed using vision alone. **Memory**: persistent file-based memory improves its performance 3× more than for Opus 4.8. **Life sciences**: with Mythos 5, Anthropic's protein design experts accelerated the process roughly 10×; 9 of 14 protein targets produced strong candidates. Mythos 5 is the first model to generate novel, compelling scientific hypotheses, preferred ~80% of the time in blind comparison; it also conducted autonomous genomics research, training a model 100× smaller that outperforms a recent publication in Science. Automated alignment evaluation places Mythos 5's misaligned behavior at a low level, similar to Opus 4.8. Customer testimonials (Cursor, GitHub, Vercel, EvolutionaryScale) confirm autonomy on long-horizon tasks and reasoning superior to Opus 4.8.

Tokenomics foundation : l'ère du FinOps appliqué à l'IA est officiellement ouverte

Thu, 04 Jun 2026 00:00:00 GMT

Analysis by **Olivier Rafal** for **WeNvision** (French consulting firm), published on **June 4, 2026** (~4 min read), commenting on the launch of the **Tokenomics Foundation** by the **Linux Foundation** (announced June 3, in partnership with the **FinOps Foundation**), which the author sees as the official opening of **the era of "FinOps for AI."** **Pivot thesis**: AI has transformed the economics of software development; the **token** has become *"the new unit of measurement for technology spending,"* mirroring the cloud of the 2010s (**recurring and variable** costs requiring active management), hence the shift by vendors from flat-rate pricing to **token-based billing**. **Order of magnitude (urgency)**: *"According to Goldman Sachs, global token usage is expected to increase 24-fold by 2030, reaching 120 quadrillion tokens per month"* — which elevates token efficiency from a *"technical detail"* to a topic for the **executive committee**. A quote from **J.R. Storment** (founder of the FinOps Foundation): *"Token costs and efficiency have become a CEO-level concern, not a technical footnote."* **Transparency/standardization problem**: current AI pricing is not comparable across providers (input tokens / caching systems / output tokens differ from one model to another) → the Tokenomics Foundation aims to **extend the open source FOCUS specification** to provide a **common language** for purchasing and comparison. **Rafal's central message (beyond cost)**: *"The point of FinOps is not so much to cut costs as to optimize efficiency"* — the real metric is **AI cost measured against business impact** (*time to market, quality, features, eco-design*). **Limits of standards alone**: technical standards are not enough — the **Target Operating Model** must be rethought (teams, processes, data culture, business alignment); American organizations are already announcing *"the end of two-pizza teams in favor of sandwich teams."* **Warning marker**: *"an AI-boosted SDLC will merely [...] amplify problems and just help you go faster... straight into a wall"* (without organizational foundations). **Cited sponsors** of the foundation: Accenture, Booking.com, Google Cloud, Microsoft, IBM, Salesforce. **WeNvision's offering**: *"co-build a roadmap, rethink the operating model for the agentic era, and establish the financial governance that has become indispensable."* **French-language, executive/transformation-oriented reading** of the [[tokenomics-foundation-linux-finops-token-economics-about-2026-06-03]] fiche; converges with the agentic FinOps cluster [[finops-foundation-finops-for-ai-overview-2026-02-17]], [[finout-finops-ai-agents-four-step-allocation-framework-2026-04-27]], [[gupta-token-budget-wars-marginal-token-utility-2026-05-28]] (token→outcome, value > volume).

About — Tokenomics Foundation (a Linux Foundation project)

Wed, 03 Jun 2026 00:00:00 GMT

**About** page of the **tokeneconomics.com** site, presenting the **Tokenomics Foundation** — a **Linux Foundation** project announced on **June 3, 2026**, operated in **close partnership with the FinOps Foundation**. **Stated mission**: *« establish open industry standards, benchmarks, and best practices for the economics of AI infrastructure »* — linking token **production, consumption, and monetization** to **business value**. **Framing definition of tokenomics**: *« Tokenomics is not just about the cost of tokens, it's about the entire layer of AI that they drive from production, to consumption to monetization »* — that is, **the entire economic layer of AI**, from infrastructure cost through model selection to value optimization. **Phase thesis**: early AI adoption prioritized **capability**; the current phase is shifting toward **efficiency and value**, which requires systematic cost management and **visibility**. **5 founding principles**: (1) ***« Efficiency is a design choice. AI cost is shaped by architecture, not just usage »*** ; (2) ***« Bigger is not always better. The best AI system is not always the one using the most expensive model »*** (right-tool / routing) ; (3) ***« Visibility comes before optimisation. Teams cannot manage what they cannot see »*** ; (4) ***« Value matters more than volume. More tokens, more calls, and more automation do not automatically mean better outcomes »*** ; (5) ***« Open knowledge benefits everyone »*** (shared standards, community learning, transparency). **Governance**: a **Governing Board** (industry direction + fund deployment) and a **Technical Committee** (open specifications + benchmarks). **Deliverables**: extension of the **FOCUS specification** (FinOps), open specs, benchmarks, frameworks and shared metrics. **Target audience**: CAIO, CTO, CIO, CFO, engineers, product teams, FinOps practitioners, researchers, startups, enterprises, public sector. **Stated goal**: moving organizations *« from experimental AI adoption to sustainable AI operations »* by extending the discipline of **variable technology spend** into the token era. **Relevance to the watch**: institutionalization/standardization of **agentic FinOps** at the level of an industry foundation — converges head-on with the notes [[finops-foundation-finops-for-ai-overview-2026-02-17]], [[finout-finops-ai-agents-four-step-allocation-framework-2026-04-27]], [[orq-ai-finops-ai-agents-cost-per-outcome-hosseini-2026-04-15]], [[gupta-token-budget-wars-marginal-token-utility-2026-05-28]] (allocation layer, token-to-outcome) and with the **token → outcome** shift (Salesforce/Tallapragada, Sierra/Greenwald). The 5 principles map exactly onto the levers already captured: architecture > usage, **Haiku/Sonnet/Opus routing**, observability before optimization, value ≠ volume.

Elon Musk Promises. Here's How Often He Delivers.

Tue, 02 Jun 2026 00:00:00 GMT

As SpaceX prepares to complete one of the largest initial public offerings in history (ticker SPCX on the Nasdaq, ~$135/share, targeted valuation on the order of $1.75 to 1.8 trillion), The New York Times publishes an interactive analysis of a decisive and rarely quantified angle: the reliability of Elon Musk's word. The promise being sold to the market is immense — a Martian colony of one million people, football-field-sized data centers in orbit, dominance in the AI race monetized against OpenAI and Anthropic. All of it rests on the leader's credibility in delivering on his commitments. The NYT compiled and coded a corpus of more than **600 dated, quantified predictions and public commitments**, made over the years in statements, social media posts, and investor calls. Verdict: fewer than one in five — about **19%** — was delivered as promised, on time or even at all. More troubling, the trend is worsening. In **2015**, Musk met nearly three-quarters of his announced goals; by **2020**, less than half were met on time, with some still awaiting their deadline years later. Two themes account for most of the repeated and postponed targets. **Mars**, cited ~19 times: a 10-20 year horizon in 2011, humans "by 2025" promised in 2016 (not realized), then Starship "within 5 years" in 2024, walked back to "by the end of next year." Then **autonomy**: more than 60 goals tied to full self-driving and the robotaxi, including the 2025 promise of a fully autonomous Tesla robotaxi, not kept. Another emblematic example: Twitter/X advertising, promised to triple, actually fell by about a third. The analysis echoes an internal admission: SpaceX's prospectus acknowledges that it is currently impossible to determine the timeline or feasibility of several major undertakings, since the necessary technologies do not yet exist. This is compounded by a narrative shift — in February 2026, Musk merged xAI into SpaceX, shifting the company's driving narrative toward AI, beyond launches and Starlink. The article's value lies not in commentary but in method: a systematic coding process that turns a stream of spectacular promises into a measurable delivery rate. It gives investors — particularly retail investors being courted — a framework for distinguishing the sales narrative from actual delivery, just as they are asked to fund the vision.

Token Budget Wars

Thu, 28 May 2026 00:00:00 GMT

Viral X thread (**230.5K views**, May 28, 2026, 1:51 AM) by **Jaya Gupta** (@JayaGup10, investor — likely Foundation Capital, author of the *Context Graphs* framework) titled ***"Token Budget Wars"***. **Pivot thesis**: ***"Enterprise AI has moved from adoption to allocation"*** — phase 1 of enterprise AI proved that models can work; phase 2 will decide **how much of that work is worth it**. The new currency at the top of the enterprise is the **ability to quantify AI ROI**: *"show me the value"*. Canonical concept: ***marginal token utility*** = *"the business value created by each additional dollar of inference"* — the number that matters at scale, and that **most companies cannot see**. Timeline: **Claude shipped November 2025**, after the 2026 annual budgets were locked → as early as **Q1**, companies *"running multiples ahead of plan"* → inference stops being an experimentation line item and becomes a **recurring operating cost**. Shift from **experimentation (a few $100K) → infrastructure (seven figures, $1M+)**: at infrastructure scale, **technical variance produces material P&L swings — two runs of the same workflow on the same input can differ by 5-10× in token cost** with nothing visibly broken, *"a number the CFO has to explain to the CEO"*. **AI competes with labor**: 3 types of budget requests (replace outsourced work / replace internal work / generate revenue) → shift toward the ***cost of a completed outcome*** (cost per resolved ticket, processed claim, reviewed contract, completed invoice, avoided hire, retained customer, dollar of revenue moved). **BPO = the easiest baseline to benchmark against** (already priced in completed units); internal work is much harder (multi-skilled employees, diffuse gains, HR resistance to headcount reduction). **Why it's different from SaaS**: SaaS learned to treat usage as a proxy for value; AI breaks that proxy — *"the signal and the noise share the same unit"* (the token), *"SaaS usage told you the software had been adopted. AI usage tells you the meter is running. It doesn't tell you whether your company is cooking."* **Three causes of marginal token utility's invisibility**: (1) ***retry tails*** — tokens per resolved workflow ≈ **T/p**; going from 90% to 70% completion increases effective cost by ~**28%**, not 20%, because failures compound; (2) ***context inflation*** — inference cost ≈ **O(n²)** in context length (attention), doubling the context **quadruples** reasoning cost (over-retrieval: 50 docs when 5 would do); (3) ***routing*** — by default the most powerful model is used (basic classification run on a complex reasoning model); across millions of calls, the difference between routing easy tasks to a small model and sending everything to the frontier model = *"the difference between a manageable bill and a board-level problem."* **Sector split**: **software** companies = a **productivity measurement** problem (already instrumented: PRs, commits, deploys, incidents, cycle time, MTTR — tracks *"AI layoffs"*); **non-software** companies = a **transformation** problem (operational work: claims, underwriting, support, compliance reviews, supply chain exceptions, payment disputes — *right under audit, not just right on average*). **The missing layer = token-to-outcome attribution**: a conversion layer linking inference spend → work performed → business outcome, answering 3 questions (real cost including retries/corrections; which parts of the trace mattered vs. thrashing; did the work change the operating model). ***Measurement becomes memory***: linking a token to an outcome requires capturing **decision traces** (what the agent saw, retrieved, called, ignored, where it retried, when a human overrode it) — *"decision rationale is one of the most perishable assets in a company"* (lives in Slack, emails, escalation calls, people's heads). Agents **create** these traces; captured first to justify the spend, they become *"more valuable than the cost report"* → a **context graph** (*"although I am so tired of that word these days"*). **The allocation layer is the prize**: whoever owns token-to-outcome attribution makes the **allocation calls** (which workflows deserve more compute, which are capped, which move to cheaper models, which stay human, which replace BPO). Companies won't do this on their own — they'll **buy it as a transformation** (Fortune 500 playbook: McKinsey + Palantir alumni + top-down CEO, in the manner of ERP/BI/digital transformation, a *"program"* with an executive sponsor and infrastructure that becomes the **new source of truth**). Framed by **Charlie Munger**: *"show me the incentive and I will show you the outcome."* Organizational sub-thesis: the decades-old executive instinct that *big teams = big jobs/scope/power* → once intelligence becomes the **scarce resource**, the new marker is *"how much of it you're orchestrating."* Direct relevance to the **Cost Optimization / agentic FinOps positioning**: empirically confirms the levers (model routing, prompt caching, context hygiene, sub-agents) and shifts the KPI toward **cost per completed outcome**. Strong convergence with Bain's *cross-system labor* (execution data moat, Cursor), Ng's *No AI jobpocalypse* (pricing anchored on the replaced employee's salary), DORA ROI (cost per feature), Mensch/Mistral (electron→token), Ensarguet (economics of computation), Foundation Capital's *Context Graphs* (decision traces, same author), Wescale's *Token Burning*, BFM/Girard (token = value fuel).

Arthur Mensch (MistralAI) devant la commission d'enquête sur les vulnérabilités numériques — compte de l'Assemblée nationale

Wed, 13 May 2026 00:00:00 GMT

**Arthur Mensch** (CEO **Mistral AI**) is testifying under oath before the **commission d'enquête sur les vulnérabilités numériques** of the National Assembly (chaired by Philippe Latombe, absent). In May 2026, Mistral has **1,000 employees**, is valued at **€12 billion**, targets **€1 billion in revenue** by end of 2026, invests **€1 billion in R&D**, with 30% of revenue in France, 70% outside France, 75% in Europe. Clients: DINUM, Caisse des dépôts, France Travail, ministère des Armées, Stellantis, TotalEnergies, BNP Paribas, Luxembourg. **Pivot thesis**: ***"cloud is artificial intelligence"*** — no distinction between digital services and AI. **Framing metaphor**: AI is a **natural resource** — *"we transform electricity into intelligence, into token generation."* **Base economics**: 1 GW of datacenter = **$50 billion in investment over 5 years**, generates **$20 billion in tokens/year** ≈ 50% gross margin; along the electron→token chain, ~10% of the value is in the electron, ~90% elsewhere. **Alarmist macro thesis**: if Europe imports 10% of its payroll in non-European AI, **an additional €1 trillion trade deficit**; **$20 trillion in infrastructure investment** is needed to serve 400 GW across Europe. ***"We don't have time"***: a **2-year** window before European energy resources are monopolized by US hyperscalers deploying **$1 trillion/year**. **Sovereignty strategy**: ***"don't think of sovereignty as isolationism but as leverage."*** Four risks: economic security (cut-off access), defense (Russian AI drones → conventional deterrence), cultural shaping (US/China biases injected), trade deficit ×5. **Defense doctrine (implicitly anti-Anthropic-Mythos)**: Mistral works with the ministère des Armées and French allies, but ***"we don't claim to have the democratic legitimacy to explain to the French armed forces what they can do."*** Duty of advice on **reliability**, not veto power over **final use**. On cyber, Mensch denounces the *"fear marketing"* of an American competitor: the offensive capabilities of models are rising *"in a linear, predictable way, for everyone at the same time."* **Campus IA** (Saint-Arnoult, €35 billion, MGX/Abu Dhabi + Nvidia, 100 hectares, 1.4–1.6 GW): Mistral is a **very minority** shareholder, potential supplier. ADEME life-cycle assessment for the models, anti-carbon-offset stance. **Regulation**: 27 unsynchronized regulations + GDPR + AI Act = ***"regulation favors the big players,"*** entrepreneurs leaving for the US. *"It's a form of colonialism"* (on the US narrative devaluing EU regulation, internalized by Europeans). **Public procurement = leverage (50% of EU GDP)**: *"the United States and China have used it massively since the 1940s — we need to stop being afraid to use it."* **Distillation = internal cost reduction, NOT technological catch-up** — so you still need to know how to train large models, which requires a lot of R&D. **Mistral's internal productivity**: ×2 in 6 months, *"Mistral engineers no longer write lines of code,"* a new posture as **agent manager**. **No bubble** on the demand side, but a **supply bottleneck** in chips/electrons. **Warning conclusion**: ***"if we combine AI strength with electrical capacity, we can regain a sustainable market share. We absolutely must do it, because otherwise we will become a vassal state."***

AI/works™ by Thoughtworks — Thoughtworks' Agentic Development Platform / "We are doing it again for the AI era"

Tue, 12 May 2026 00:00:00 GMT

Thoughtworks launches **AI/works™**, its **agentic development platform** claimed to be *"the new standard for building and running industrial-grade systems in the AI era"*. The opening tagline — ***"We are doing it again for the AI era"*** — explicitly invokes Thoughtworks' heritage (XP, Continuous Delivery, microservices, refactoring) to sell the new platform. The **core thesis is economic**: *"The breakthrough is the economics. The old approach made you pay millions to build, run, then pay again to rebuild. AI/works™ ends that routine."* Four derived promises: continuous system updates, selective regeneration *"without the token blowout"*, *"your systems finally stop aging"*, and fast-tracked new products. The platform deploys **six capabilities** covering the entire SDLC: (1) **Reverse Engineering** ingests legacy code and produces validated as-is specs; (2) **Dynamic Spec Development** converts raw requirements into a unified *Super Spec* covering architecture, workflows, security, data, UX; (3) **Spec to Code** generates testable code from the Super Spec via *coordinated agents*; (4) **Developer Experience** standardizes AI-assisted development via *governed golden paths*, automated pipelines and a shared catalog; (5) **Control Plane** orchestrates and governs agents with *cost transparency*, *active guardrails* and *end-to-end lineage*; (6) **Runtime Ops** continuously monitors, detects change, updates the Super Spec and regenerates impacted code. The pivot concept is the **Super Spec**: a **dynamic, unified specification** serving as the source of truth, automatically updated in production and triggering regeneration of impacted code rather than patches. The **3-3-3 methodology** structures delivery: 3 days to align the product concept, 3 weeks for a prototype (desirability/viability/feasibility), 3 months for a production MVP. ***"Industrial-grade systems that grow up instead of grow old."*** **Constellation Research** recognizes AI/works™ *"for changing the economics of enterprise software delivery"* via a *"spec-driven, lifecycle"* approach. The **anti-positioning** is explicit: *"no consultant crowds"* (a direct jab at the large integrators), *"finance can open the bill without switching on emergency lighting"* (corporate self-deprecation), *"stands on an engineering foundation rather than enthusiasm"* (deliberate anti-hype). The **featured partners** span the entire agentic stack: AWS, GCP, Azure, Databricks, Snowflake (cloud/data), Claude, DeepSeek, Gemini, Grok, OpenAI (LLMs), NVIDIA, Groq (compute), CAST, Mechanical Orchard (legacy), Stripe, Spotify (likely reference customers). The **dual CTA** *Request a discovery call / Sign up for updates* betrays a **sales-led, high-ACV** model. Read within the 2025-2026 corpus, AI/works™ is the **productization** of the Thoughtworks doctrine carried intellectually by Kamelman (*Service-as-Software*, 2025-12), Fowler (*LLM Retreat*, 2026-02) and Böckeler (*Harness Engineering*, 2026-04). It is the **Anglo-Saxon commercial equivalent** of the Wescale doctrine *Usine Logicielle Augmentée* (2026-05-03), packaged as a platform.

The Batch n°352 — "There Will Be No AI Jobpocalypse" (Andrew Ng)

Fri, 08 May 2026 00:00:00 GMT

**Issue 352 of The Batch**, DeepLearning.AI's weekly newsletter published on **May 8, 2026**, opens with an editorial by **Andrew Ng** titled **"There Will Be No AI Jobpocalypse"**. Ng dismantles the narrative of mass unemployment caused by AI, drawing on macro data: **healthy 4.3%** US unemployment rate, robust tech hiring despite major progress in software engineering. Rather than refuting the narrative with figures alone, Ng identifies **three structural drivers** of the jobpocalypse narrative. **First driver — tech incentives**: AI labs benefit from presenting their technology as transformative-disruptive. They raise more funding, attract more talent, see their valuations climb. The more credible the fear of replacement, the more justified the value attributed to the models appears. **Second driver — pricing power**: enterprise AI vendors charge **$10,000+ per year** to their clients by anchoring their pricing on the **salary** of the employee their product is supposed to replace, rather than on traditional SaaS pricing (per seat / per usage). This is the **service-as-software** shift in its financial version: if the product "replaces an $80,000/year employee," $20,000/year seems reasonable. **Third driver — corporate messaging**: companies reframe their layoffs as *"AI efficiency"* rather than acknowledging the **pandemic-era overhiring** of 2020-2022. This narrative is sellable to markets and the public, whereas admitting a prior strategic mistake is uncomfortable. Ng honestly acknowledges: *"AI disrupts work"*. But he **flips the narrative** by proposing the neologism **"AI jobapalooza"** (a play on Lollapalooza, festival → abundance). The substance: job creation in AI engineering and adjacent fields, with evolving skill sets. The editorial fits into a **contrarian series** characteristic of Ng: dismantling hype cycles, defending engineering pragmatism against grandiose announcements. The implicit target is **Dario Amodei** (Anthropic) and his prediction of **50% of white-collar jobs eliminated by 2030** — Ng points out, without naming him, that **Anthropic benefits from promoting this narrative**. The timing is striking: the editorial appears **the same day** that **David Wallace-Wells** publishes his sprawling NYT Magazine piece on **"AI Populism"** and the anti-tech backlash (the Altman Molotov cocktail incident, the Indianapolis shooting). A perfect mirror reading: Ng conducts a **cold economic analysis** of narrative incentives, while Wallace-Wells documents the **emotional popular panic** they fuel. The stakes for Ng are not merely intellectual: protecting decision-makers and workers from hasty decisions (preemptive layoffs, panic, individual despair) triggered by a narrative that primarily serves AI vendors.

The $100-Billion SaaS Opportunity Hiding in Cross-System Labor

Fri, 01 May 2026 00:00:00 GMT

**Bain & Company** publishes in May 2026 (David Crawford, Chris McLaughlin, Greg Fiore) a brief, part 2/5 of a series on *"the software industry in the age of AI"*. **Pivot thesis**: the major opportunity in agentic AI **is not to replace existing SaaS** but to **automate cross-system coordination labor** — *"employees pulling budget data from an ERP, checking inventory in a spreadsheet, interpreting free-text responses, and making judgment calls"*. **Market sizing**: ~$100B in the US (~$200B including Canada/Europe/AU/NZ). **Current capture $4-6B (10%)** — so **>90% still up for grabs**. US distribution: Sales ($20B) + COGS/ops ($26B) + R&D ($6-12B) + support ($6-12B) + finance ($6-12B). **Six automation factors** to assess a workflow: (1) output verifiability, (2) consequence of failure, (3) digitized knowledge availability, (4) integration complexity, (5) process variability, (6) physical world dependency. **Potential by function**: Customer support & R&D **40-60%**, Finance & HR **35-45%**, Sales & IT **30-40%**, Legal **20-30%**. **Strategic shift**: competitive advantage moves from *system of record ownership* (Salesforce/SAP/Workday) to ***cross-workflow decision context*** — the cross-cutting ability to see and act across multiple integrated systems. **Durable moat**: ***"accumulated execution data that grows more valuable over time and becomes harder for competitors to replicate"***. **Four examples**: **Sierra** (autonomous customer issue resolution), **Glean** (cross-function employee request coordination), **GitHub Copilot** (extended beyond source control), **Cursor** (ARR doubled in a quarter, reaching $2B). **Three-phase playbook**: (1) **Assessment** — six factors + market sizing; (2) **Strategic Positioning** — data assets + adjacent workflows + actual operational maps; (3) **Execution** — build/buy/partner + restructure org + ***redesign data foundations for agent readiness***. **Dossier connections**: strong convergence with **DORA ROI 2026** (ROI financial framework), **Foundation Capital Context Graphs** (decision traces), **Seale Semantic Agent** (ontology as moat), **Habert PROJ-AI** (six zones + doctrine), **Talisman Ontology Pipeline Refresh** (governance + AI partnership). Productive tension with **MIT NANDA 95% pilots fail**: the two converge — pilots fail precisely because 90% of the market remains unstructured. **Sierra** appears in 3 notes in the dossier (Bain as reference case + 2 AI-native interview notes), confirming its emblematic position. To be used for SaaS executive committees / PE / VC due diligence / CDO data foundations.

Silicon Valley Is Bracing for a Permanent Underclass

Thu, 30 Apr 2026 00:00:00 GMT

Jasmine Sun publishes a long investigation in NYT Opinion (April 30, 2026) documenting Silicon Valley's silent fear of the emergence of a *permanent underclass* — a viral theory holding that AI could freeze economic mobility and render millions of people economically useless. The *San Francisco consensus*, shared across the board (engineers, VCs, doomers, lefties), boils down to one sentence: *"the median person is screwed, and they have no idea what to do about it."* The risk has shifted from the dystopian AI register (rogue AGI) to the mundane register: mass elimination of white-collar jobs. Dario Amodei (Anthropic) publicly predicts *"a white-collar blood bath"* and 50% of junior white-collar jobs disappearing by 2030. Sam Altman had already warned in 2021 about the labor → capital shift. Block (Jack Dorsey) lays off half of its workforce in March 2026, explicitly citing Opus 4.6 and Codex 5.3 — the stock market responds with a +25% surge. OpenAI's **GDPVal** benchmark measures 44 occupations and reaches *"over 80% win rate compared to human professionals"* within a few months. Sun identifies a dissonance between public statements and actions. In April 2026, OpenAI publishes a white paper, *"Industrial Policy for the Intelligence Age,"* with radical proposals (32-hour workweek, public wealth fund, higher capital taxation), but no concrete legislative commitment. The pro-AI PAC *Leading the Future* (partly funded by Greg Brockman) spends more than $2M against NY candidate Alex Bores, who proposes safety regulation. Anthropic opens the **Anthropic Institute** (March 2026, led by Jack Clark) with ARR exploding to $30B versus $9B at end of 2025; a $20M political contribution counterbalances in favor of Bores, but there is still no economic policy paper. Among economists: Korinek (UVA) — *"no human job is invulnerable"*; Autor (MIT) — new occupations will emerge; Frey (Oxford) delivers the epitaph: ***"the short run can be a lifetime."*** Political strategy crystallizes around David Shor: 79% of voters are worried about the absence of a government plan, the *federal jobs guarantee* tests better than UBI, and the slogan *"They work for the bots. We work for you."* dominates ad testing. The 2028 presidential race will be politically structured by AI. Mark Kelly and Ro Khanna announce sweeping AI agendas. April 2026: an attempted firebombing of Altman's home, an attack on a pro-data-center Indianapolis councilman. Alex Karp (Palantir) warns his peers: *"the country could blow up politically and none of us are going to make any money."* Sun's political thesis: the creation of an underclass is a **policy choice**, not a technological inevitability. The moment is open for radical redistributive policies — provided action comes before populist frustration turns to violence.

FinOps for AI Agents: A Four-Step Allocation Framework

Mon, 27 Apr 2026 00:00:00 GMT

Finout proposes an operational framework for allocating AI agent costs — a problem distinct from cloud FinOps. The scope covers coding assistants (Claude Code, Cursor, GitHub Copilot), agents embedded in customer-facing products, and direct LLM API spend (Anthropic, OpenAI). The starting observation: Finance teams receive from AI vendors *"a single line-item bill they cannot allocate to the responsible cost centers,"* an opaque, fast-growing shared cost that prevents tracking unit economics, team-level accountability, and the COGS of AI features. The article identifies **three structural properties** that invalidate cloud FinOps assumptions. (1) **Per-call cost is non-deterministic**: the same prompt issued by two developers produces different bills depending on context length, retries, agentic loop depth, and model variant. (2) **There is no taggable resource at the point of use**: using Cursor does not provision any cloud resource carrying metadata. (3) **Consumption does not map to environments**: refactoring an internal service or building a customer-facing feature costs the same, yet their business value differs. Notable figure: a developer working in greenfield mode consumes **5 to 10× the tokens** of a developer doing code review — which is why per-head chargeback fails. This gives rise to **four allocation problems**: per-developer attribution of IDE assistants; embedded-feature spend that must be treated as product COGS; cost-per-customer / per-feature / per-tenant calculations; and shared spend with no tagging at the source. The core of the article is a **four-step framework**: (1) centralize vendor bills as first-class sources normalized alongside cloud spend; (2) replace source-level tagging with **rules-based allocation** expressed in the team taxonomy, with the logic hosted inside the FinOps system itself; (3) link agent activity to **identity** (SSO, API key, seat) correlated with HR systems, making allocation automatic and resilient to role changes; (4) treat embedded-agent spend as **product COGS**, in the same bucket as infrastructure. Guiding principle: the platform must support **allocation logic that the FinOps team can edit without engineering involvement**, since AI spend is *"among the most volatile line items"* in the tech stack (new models monthly, quarterly reorgs). Finout finally positions its building blocks — MegaBill (ingestion), Virtual Tags (ownership without source tagging), Unit Economics, back-allocation of shared costs — as the tooled response to the agentic era.

FinOps for AI Agents: How Enterprises Control Cost, Value, and Scale

Wed, 15 Apr 2026 00:00:00 GMT

Sohrab Hosseini (co-founder of Orq.ai) argues that traditional FinOps — designed for deterministic workloads with infrastructure-driven costs — is structurally insufficient for AI agents, whose spend depends on **runtime behavior**. A single user request can trigger variable sequences (retrieval, tool calls, model routing, retries, escalations), such that *"two requests that appear identical to a user can produce very different token usage"* with no visible change in functionality. The diagnosis is backed by statistics: **80%** of enterprises use GenAI in 2026, but **less than 30%** have monitoring that links cost to value; only **27%** allocate cloud costs in real time, **less than 25%** have standardized AI governance, and **74%** struggle to industrialize their pilots. The proposed response is a conceptual shift toward **"cost per outcome"**: measuring cost per ticket resolved, lead qualified, task completed, or hour saved — not tokens or infrastructure usage. The relevant question is no longer "how many resources" but *"whether the consumption produced value"*: a token-efficient agent that fails costs more than a token-hungry agent that succeeds at a complex task. To achieve this, **Agent FinOps** integrates **three layers of signals**: **cost** signals (model usage, tokens, API spend, budgets), **operational** signals (traces, retries, routing decisions, evaluation results), and **business** signals (resolution rate, completion, conversion, time saved). This integration unfolds across a **four-phase lifecycle**. *Experiment*: bounded experimentation with budgets, cost-per-evaluation, and unit economics established before production. *Deploy*: guarded releases with routing policies, token limits, and timeouts ensuring economically predictable behavior. *Operate*: visibility into retries, escalations, and consumption to proactively adjust routing and constraints. *Improve*: evaluations guide prompt refinement, workflow redesign, model selection, and the retirement of underperforming automations. Operational levers — **guardrails**, intelligent **model routing**, workflow budgeting, context control, and **behavioral observability** — converge into a centralized layer, the **Control Tower** (unified agent inventory, cost rollups, governance). Marker phrases: *"A single agent is a feature. A collection of agents becomes an operational environment"* and *"Enterprises aren't struggling because they can't build agents. They're struggling because they can't coordinate them."* The final principle: agentic FinOps does not scale by tracking spend more aggressively, but *"by shaping agent behavior at every stage."*

Starving Genies

Fri, 03 Apr 2026 00:00:00 GMT

In "Starving Genies," Kent Beck analyzes the usage limits recently imposed by major AI service providers through the lens of growth economics. He uses the concept of the "Expand phase" — drawn from the Explore/Expand/Extract model — to explain that a company's growth follows a staircase rather than a curve: progress continues until it hits a limiting resource, at which point the company must either increase the supply of that resource or reduce demand to unlock the next step. Beck observes that Google, Amazon, and Anthropic all implemented usage limits in near-simultaneous fashion, which in his view reflects a signal sent to investors and a deliberate economic narrative choice more than a simple unforeseen technical constraint. As a daily practitioner of augmented coding — he works with his "genies" (AI assistants) all day — Beck describes the concrete impact of these limits. A power user who hits their cap mid-workflow does not experience a minor inconvenience: their work stops dead. These limits segment the user base into three categories: casual users get free but capped access; developers access the API with metered pricing but no abrupt daily cliff; and in between, technical power users who are not API-oriented end up pushed toward premium consumer subscriptions. This is precisely the conversion pressure companies are looking for. Beck then raises an unsettling question: are these limits a temporary bridge — until compute supply catches up with demand — or do they mark the beginning of a new equilibrium in which intensive AI usage becomes a premium product rather than a default service? On the technical and economic side, he notes that inference is progressively becoming cheaper through Distillation de modèles, intelligent caching, routing to smaller models based on query complexity, and the maturation of custom silicon. Competitive dynamics are crucial: a company that manages to significantly cut its unit inference costs can afford to open the floodgates while competitors ration usage, thereby capturing the next wave of users. Beck points out that Nvidia H100 GPUs remain scarce and contested — with the notable exception of Google, which manufactures its own TPU chips and therefore holds a structural advantage in this capacity race. The article closes on a gentle irony: Beck says he "totally believes" that Anthropic hit a physical wall overnight, while hinting that the reality is more nuanced.

AI Brings Headwinds and Tailwinds to the Rule of 40

Wed, 01 Apr 2026 00:00:00 GMT

**Bain & Company** publishes in April 2026 (David Lipman, Greg Callahan, Daniel Goetz, George Sunderland) a part **1/5** brief in the series *"software industry in the age of AI"*, dedicated to AI's impact on the **Rule of 40** (canonical SaaS metric: *growth rate + profit margin ≥ 40%*). **Pivot thesis**: the Rule of 40 is under **dual pressure** — *headwinds* (slowing market growth + massive AI infrastructure costs) and *tailwinds* (AI productivity + 10-25% EBITDA transformation + outcome-based pricing). SaaS leaders may have to ***"settle for the Rule of 30"*** temporarily to remain competitive against AI-natives. **Striking data point**: a *marketing technology* client case — **AI costs multiplied by 3.49 (+349%)** while **revenue increased only 38%** over one year. Illustrates how AI infrastructure + inference + model access introduce **significant variable costs** into businesses that historically ran on **high fixed margins**. **Headwinds**: (1) *"software penetration is topping out in some areas"* — saturation of mature markets; (2) AI variable costs compress margin. **Tailwinds**: (1) sales/marketing/R&D productivity; (2) **successful transformations = +10-25% EBITDA**; (3) *outcome-based pricing* opportunity (revenue shifted from *fixed seats* toward *labor/operations economics*); (4) incumbents can leverage customer relationships and embedded workflows against AI-native challengers. **Two paths forward**: (1) ***Financialize*** — minimize AI investment, optimize cash, operate as a *"durable generator"* but limit future growth; (2) ***Invest to Grow*** — accept short-term margin pressure, reinvest aggressively in AI across product and operations. **Watch-file connections**: convergence with **Bain part 2/5** *Cross-system labor $100B* (May 2026), **DORA ROI 2026** (verification tax / instability tax), **Cherny Sequoia** (7 Powers reordering), **Menlo Ventures** (State of Generative AI Enterprise), **Foundation Capital Context Graphs**. Productive tension with **MIT NANDA 95% pilots fail** / **DORA market divide**: there are two populations of SaaS in 2026 — those that transform (Rule of 30 → Rule of 40+) and those that stagnate. *Outcome-based pricing* convergence with **Levie** (*Building for trillions of agents*) and **Sierra** (autonomous resolution) — the per-seat SaaS model is under pressure to be replaced by outcome/consumption/labor-substitution models. Relevant for SaaS CFOs/boards (budget framework), PE/VC investors (headwinds/tailwinds due diligence grid), SaaS CEOs (Invest to Grow argument with +10-25% EBITDA), executive committee presentations (the +349% / +38% case as an alert on uncontrolled AI cost).

How to Use AI for Market Research (Step-by-Step Guide) — AI Market Research Tension Map

Mon, 30 Mar 2026 00:00:00 GMT

Alex Pawlowski publishes in *The Strategy Stack* a method guide that repositions *market research* for the AI era. His diagnosis opens with three flaws: information overabundance that creates noise, delayed reports describing the past, and insight rarely translated into action. His thesis: value no longer lies in the static report but in a **living decision surface** — an evolving model maintained as an operational system. Markets are treated there as *dynamic strain-fields*, not as fixed competitive landscapes. His central contribution is the **Tension Map**: instead of mapping competitors and market share, it identifies contradictions and pressure points — *"where users want more than products deliver"*, price/value misalignment, powerful incumbents lacking emotional resonance, friction accepted for lack of an alternative. The Tension Map reveals *opportunity spaces* invisible to classic analyses. Three research modes structure the method: **Discovery Mode** establishes the baseline (players, visible patterns); **Tension Mode** locates dissatisfaction and underserved segments; **Decision Mode** converts interpretation into action. The **7-step workflow** operationalizes this: (1) define a precise question, (2) collect *raw signals* (reviews, docs, transcripts — not synthesized reports), (3) build the Tension Map, (4) AI-driven structural analysis, (5) stress-test through adversarial perspectives, (6) convert tensions into decisions, (7) preserve as an *updatable market model*. The tool stack is orchestrated by phase: **Perplexity** for expansion (Discovery), **Claude** for depth and continuity (large context, contradictory signals), **ChatGPT** for iteration speed (reframing, alternative structures), **multi-agent** for *productive disagreement* via assigned roles (analyst, critic, strategist). Question strength is central — *"Where does pricing feel tolerated rather than embraced?"* (strong) vs *"What are the trends?"* (weak). Pawlowski illustrates this with the AI note-taking tools market: the automation promise is appreciated but post-meeting accuracy is problematic, premium pricing is tolerated in teams but resented individually, incumbent trust competes against emerging excitement. **Stress-testing** is a ritual step: *"what would invalidate the interpretation?"*, *"what would skeptical competitors dispute?"* Four failure modes loom: vague questions producing polished but superficial outputs, over-reliance on *polished summaries* at the expense of raw signals, skipped validation creating ungrounded confidence, insights never translated into decisions. **Corpus persistence** in Claude Projects and **living model maintenance** turn research into a compounding asset. Methodological reference: Richards Heuer's (CIA) *Analysis of Competing Hypotheses*, transposed to AI. Market research becomes a continuous operational system, not a succession of discrete projects.

The Age of Open Agentic Commerce

Thu, 19 Mar 2026 00:00:00 GMT

Sam Ragsdale argues in this essay that true agentic commerce will be open and permissionless, built on simple protocols, rather than closed platforms like ACP (OpenAI/Stripe) and UCP (Google), which he sees as the equivalent of AOL in the 90s — a walled garden with better UX, but a dead end for innovation. The article builds its argument on a detailed historical parallel. In the 90s, two visions of the internet competed: AOL (curated content, flat pricing) versus open protocols (HTTP, DNS, HTML, Mosaic). AOL seemed to have won with its $350 billion Time Warner merger, but open protocols ultimately enabled the emergence of Facebook, Google, and Amazon — innovations that came from the edges, without permission from gatekeepers. The internet's economic model was built on a hack: advertising. In 1997, Tim Berners-Lee created the HTTP 402 (Payment Required) code to enable micropayments, but fixed credit card fees made one-cent transactions impossible. Google worked around the problem by monetizing attention via advertisers. But AI agents fundamentally change this equation: they are not distractible. StackOverflow has lost 75% of its views since GPT-4, and tech site traffic has dropped 60%. Even walled gardens like Facebook and TikTok are now being breached by computer-use agents that perfectly mimic human behavior. The solution lies in two emerging protocols: x402 (Coinbase) and mpp (Tempo + Stripe). Twenty-eight years after the invention of the 402 code, stablecoins on modern blockchains offer sub-cent transaction costs, finally solving the micropayments problem. These protocols allow agents to pay for any service — data, hosting, communication — without a prior commercial agreement, without a BD process, without a whitelist. Ragsdale presents AgentCash as the missing discovery layer: a single balance giving access to all APIs, with merchant registries (x402scan.com, mppscan.com) where services register to be found by the 2,000+ agents already active. He sees skills as a transitional artifact, since modern agents (Claude 4.5+, Codex 5.2+) can discover an unfamiliar API, read its schema, and use it correctly without prior training. The ultimate vision is one of hundreds of millions of agents accessing hundreds of thousands of services autonomously, recreating the permissionless innovation dynamic that made the open web great.

Beyond Brain Speed: The Economics of Computation

Wed, 11 Mar 2026 00:00:00 GMT

Philippe Ensarguet develops a strong thesis: we are living through the "kilowatt-hour moment" of the knowledge economy. Just as the light bulb ended the era of the candle, AI ends the brain-hour as a unit of economic value. He opens with a striking anecdote: a consultant delivers three weeks of work that the client replicates in four minutes with an agent, a thousand times cheaper. For decades, the knowledge economy relied on a proxy: time was paid for (person-days, billable hours) in the hope that value would follow. This system created a systemic distortion in which slowness was rewarded. When an AI agent produces in a few minutes the code, the audit, or the report that used to justify several billable days, the very foundation of the model collapses. This is the commoditization shock of our generation, comparable to what the cloud did to infrastructure. Ensarguet invokes the Jevons paradox: when execution cost falls to near zero, demand becomes infinite. We will not expect less work, but 1000 times more output. The signals are already there: budget lines are shifting from salaries to computing power. He proposes a new multiplicative value equation: Value = Computation Quality x Data Context x Orchestration Intelligence. If any one of the three factors is zero, total value collapses. But execution represents only a third of project time. Agentic AI tackles the remaining two-thirds: meetings, alignment, approvals, coordination. This second wave calls the org chart itself into question. Coordination layers (project managers, middle management) face the same pressure as execution roles. The unit of work shifts from "a person assigned" to "an outcome contracted to a system". Ensarguet identifies three emerging human roles: trajectory guardians (strategic and ethical alignment), context shepherds (sector-specific nuance, client history, regulatory subtlety that models cannot learn on their own), and trust architects (guardrails, governance of autonomous systems). Human intelligence, which incorporates values, emotions, culture, and lived experience, remains irreducible. In conclusion, he recommends three actions: experimenting with outcome-based pricing now, tracking the salary-to-compute spending ratio as a strategic indicator, and reorganizing the company around human-agent orchestration.

Local LLMs vs Cloud APIs: 2026 Total Cost of Ownership Analysis

Thu, 05 Mar 2026 00:00:00 GMT

SitePoint presents a total cost of ownership (TCO) analysis comparing locally run LLMs against cloud APIs, looking ahead to 2026. The core thesis: comparing on per-token price alone is a trap. The sticker price on an API rate card or the MSRP of a GPU tells only a fraction of the real story; only a full TCO model, over 12 and 36 months, incorporating hardware, electricity, cooling, and labor, allows for a sound decision. The article builds this model across three usage tiers: light, medium, and heavy (10 to 100M+ tokens/day). On the cloud side, the article publishes a price-per-million-tokens grid that reveals two major asymmetries. First, **output costs 4 to 5 times input** (GPT-4.1: $2 input / $8 output; Claude 4 Sonnet: $3/$15; Claude 4 Opus: $15/$75). Second, the **gap between models reaches ~150x** on input, from GPT-4.1 nano ($0.10) to Claude 4 Opus ($15): the model — its publisher, generation, size — determines the unit cost of the token, much like the cost of electricity production depends on its source. On the local side, hardware (RTX 5090 at $1,999, a Mac M4 build at $6,150, AMD MI325X) doesn't pay off before **15 to 20M tokens/day**, and only reaches parity against the cheapest hosted options at 36 months under sustained heavy usage, for an effective cost of about **$7.15/M tokens**. The costs everyone underestimates — electricity, cooling, labor (up to 30-60 hours/month at the heavy tier) — weigh heavily. Sensitivity to electricity prices is striking: moving from the US rate ($0.12/kWh) to European rates ($0.25-0.30/kWh) pushes the break-even point 40 to 60% higher in daily volume. The central takeaway, which serves as the title-conclusion: **2026 break-even points are 40% lower than in 2024**. The structural decline in hardware costs and the maturation of open-weight models make local deployment viable at increasingly accessible volumes. The final decision depends on the profile: local becomes relevant for sustained volumes, sovereignty/confidentiality requirements, and a 3-year amortization horizon; cloud retains the advantage of flexibility, no upfront capital, and access to frontier models. Beyond cost, the article also notes the performance, confidentiality, and flexibility trade-offs.

Redesigning the Agency Value Model (rapport VoxComm 95 pages, mars 2026) + Billable Hours Are Dead, AI Killed Them, Here's How To Survive (MediaPost / Joe Mandese, 3 mars 2026)

Sun, 01 Mar 2026 00:00:00 GMT

**Consolidated dossier** March 2026 on the **death of the billable hours model in the advertising/communications industry** — combining the **VoxComm report** *"Redesigning the Agency Value Model"* (95 pages, March 2026, **Brian Kessman** of **Lodestar Agency Consulting** + foreword by **Tim Williams** of **Ignition Consulting Group**, intro by **Charley Stoney** President of VoxComm / CEO of **EACA European Association of Communication Agencies**) and the **MediaPost opinion article** *"Billable Hours Are Dead, AI Killed Them, Here's How To Survive"* (March 3, 2026, **Joe Mandese**, Editor-in-Chief of MediaPost). **Shared pivot thesis**: the business model of communication agencies (billable hours / labor-based compensation / service business model) is **structurally disqualified by AI**; agencies must ***"decouple revenue and profit from staffing numbers"*** (Stoney). **MediaPost figures (Mandese)**: agency margins **30% (golden age) → 10% (current average)**; creatives produce **~5× the output** for the same pay or less than 10 years ago. **Mandese's diagnosis**: *"We are defining and monetizing our value through time and effort rather than business impact"* — when agencies sell **hourly services**, they sell **commodities** vulnerable to **AI cost compression**. **Tim Williams quote**: ***"At the heart of our industry's challenges lies a simple economic truth: incentives matter. When agencies embraced the hourly rate model, they unknowingly created a structural misalignment. What agencies are rewarded for — more hours — clients are incentivized to minimize."*** Zero-sum outcome, **race to the bottom**. **Williams' pivot solution**: ***"You are not in the service business. Agencies don't sell services and capabilities, but rather solutions to business problems."*** **Mandese's 4-shift framework**: (1) Define narrow expertise areas; (2) Codify repeatable productized solutions; (3) Build teams around outcomes, not utilization; (4) Replace rate cards with value-based models (fixed fees, subscriptions, performance-based pricing). **Concrete examples**: **FIG** (decoupled pricing from staffing), **72andSunny** (modular product menus), **Monks** (single subscription combining talent + tech + improvement). **Methodological critique**: MediaPost commenters dispute the historical 30% margin figure, suggesting real figures closer to 12-15%. **VoxComm report** structured into 8 chapters: When Your Model Works Against You / Mapping Your Value Model / Case Studies / How to Pivot / How to Price / How to Plan / How to Navigate / Online Tools. **Major relevance** for the dossier: this is the **agency counterpart** of the consulting shifts (McKinsey/Sternfels 60,000 = 40,000 humans + 20,000 agents, January 2026) and SaaS shifts (Bain Rule of 40 → Rule of 30, April 2026). **Cross-cutting convergence for knowledge-intensive services**: consulting + agencies + SaaS are simultaneously shifting from *time-and-materials* to *outcome-based*. To be leveraged for agency/consultancy/marketing/communications executive committees, strategic presentations on AI transformation of services, sourcing on the 30%→10% margin figures.

L'Agentic Commerce Optimization : le Guide technique pour se préparer aux protocoles ACP et UCP de Google

Mon, 23 Feb 2026 00:00:00 GMT

Nicolas Marette publishes a technical guide to prepare businesses for Google's Agentic Commerce Protocol (ACP) and Universal Commerce Protocol (UCP) protocols, which redefine how brands interact with AI agents in the online purchasing journey. UCP, announced in January 2026, is already operational: as of February 11, 2026, Wayfair and Etsy are using it for direct AI-driven purchases. The protocol rests on **six core capabilities**: product discovery for AI agents, cart management with complex pricing rules, identity linking via OAuth 2.0, checkout process, order management via webhooks, and vertical industry extensions. The article emphasizes that **schema.org remains the essential foundation** of this transformation. Although UCP uses JSON schemas, structured data remains the "cement binding all ontologies together," enabling AI agents to understand and navigate the e-commerce ecosystem. **Merchant Center** configuration becomes critical: mandatory return policies, customer support information, the `native_commerce` attribute for payment eligibility, and unique product identifiers. A major paradigm shift concerns **conversational attributes**: product FAQs, compatible accessories, substitute products, cross-selling opportunities, and enriched descriptions (for example "wolf" rather than a generic "purple"). This data addresses the discovery patterns of the AI era, in which agents query catalogs with unprecedented granularity. The author uses **WordLift**'s visual diffusion simulator to demonstrate how a single product image breaks down into multiple search intents, revealing the attributes that AI agents prioritize when distributing products. The **UCP roadmap** plans extensions toward multi-item carts, complex promotions, standardized loyalty management, post-purchase support, advanced personalization signals, and expansion into travel, services, digital goods, and hospitality. The article stresses the importance of **third-party social proof**: reviews from platforms such as Trustpilot, G2, or Custplace are frequently cited by LLMs and validate the integrity of product data. The conclusion is a five-step call to action: join the UCP waitlist, configure Merchant Center, train technical teams on UCP documentation, fill in conversational attributes, and audit schema implementations. Marette warns that this transformation is happening faster than previous e-commerce shifts and that brands taking a wait-and-see approach will already be behind by the time full rollout occurs.

FinOps for AI Overview

Tue, 17 Feb 2026 00:00:00 GMT

*FinOps for AI Overview* is the FinOps Foundation's reference guide, co-authored by a broad working group (Google, AWS, MetLife, Wells Fargo, Roche, Accenture, KPMG, EY…) and published under a CC BY 4.0 license. It extends the FinOps discipline to generative AI services, starting from the token as the fundamental unit of consumption, whose "meters" differ profoundly from classic cloud metrics. The document provides a battery of **KPIs with formulas and worked examples**: Cost Per Token (total cost / tokens), Cost Per Inference (inference costs / requests, e.g., $0.05), Training Cost Efficiency (cost / accuracy point), ROI ((benefits − costs)/costs × 100), and above all the *LLM Model Choice Quality Score Alignment*, which compares the minimum MMLU score a task requires against the MMLU of the model actually used, to detect over-provisioning (a sentiment analysis task requiring MMLU 54 should not run on GPT-4). On the **optimization** side, the focus is on token reduction (shortening prompts while preserving clarity), **caching** of repeated responses, **model selection** (*"avoid using the most complex and expensive models for every task"*), and **model distillation** for production. The structural core maps the **14 capabilities of the FinOps Framework** onto "common to cloud" versus "different for AI." The most affected: **Allocation** (traceability of *multi-agent workloads*, absence of a standard framework), Planning (estimating successful outputs and separating them from hallucinations), Forecasting (lower predictability in early phases), Benchmarking (per-token metrics, few external benchmarks), Unit Economics (cost-per-call, customer satisfaction per dollar), and Rate Optimization (volatile pricing such as OpenAI Scale Tier). Maturity progression follows a **Crawl → Walk → Run** model: *fail-fast* prototyping and manual calculations at the start; basic tracking automation and anomaly detection next; advanced tracking, integrated financial metrics, and vigilance against cutting costs that compromise non-functional requirements in the Run phase. The document lists **eight pricing models** (on-demand, reserved/CUD, provisioned — OpenAI Scale Tier, Azure PTU —, spot/batch, subscription, tiered, freemium, hybrid) and favors **showback** as an awareness lever ahead of chargeback. A notable limitation: **AI agents** are not yet treated as a distinct category (only *multi-agent workloads* surface under Allocation), which explains the value of vendor complements. The guide is accompanied by a *Certified FinOps for AI* certification and points to FinOps X 2026 (June, San Diego).

AI Shopping Assistant Guide 2026: Agentic Commerce Protocols

Tue, 10 Feb 2026 00:00:00 GMT

Max Thilén publishes a 29-minute reference guide on the agentic commerce protocols structuring e-commerce in 2026. The starting observation is radical: customers now complete entire purchase journeys inside AI conversations, without ever visiting merchant websites. ChatGPT Instant Checkout serves 900 million weekly users. Two protocols are competing for this infrastructure. **ACP** (Agentic Commerce Protocol), backed by OpenAI and Stripe, has been operational since September 2025. It imposes a 4% commission per transaction, a structured product feed, and 5 REST endpoints for checkout. The flow is entirely conversational: the user asks, the agent searches, displays a "Buy" button, and Stripe processes payment via a shared token — the agent never sees the banking data. **UCP** (Universal Commerce Protocol), announced by Google in January 2026 at NRF, federates a massive coalition: Shopify, Walmart, Target, Wayfair, Visa, Mastercard, PayPal, and 20 other partners. Its architecture is more modular, with four transport options (REST, MCP, A2A, Embedded) and two checkout modes (native within the AI interface or iframe to the merchant site). Merchants declare their capabilities via a `/.well-known/ucp` profile. Amazon's absence is strategically significant. Controlling 40% of US e-commerce, Amazon has chosen a proprietary path with Rufus AI (300 million users, +60% conversion), Alexa+ for voice commerce, and "Buy for Me," which allows purchasing from competitors within the Amazon app. Amazon has even blocked OpenAI's crawlers, removing 600 million products. The market thus forms a triptych: proprietary Amazon, Google UCP, OpenAI ACP. The guide details the technical requirements of both protocols, notably **product data**, which becomes the primary competitive lever when paid advertising doesn't reach agents. Descriptions must serve natural-language processing, not classic SEO: describing use cases and materials rather than stacking keywords. The **attribution crisis** is the major challenge: 70-90% of purchase journeys are already invisible to traditional analytics, and agentic commerce approaches 100%. The first measurable signal is an order webhook. Thilén recommends server-side tracking, incrementality testing, and AI visibility monitoring, with maturity estimated at 18-24 months. The projections are considerable: McKinsey anticipates $3-5 trillion in global revenue by 2030. For brands, the conclusion is pragmatic: both protocols matter. ACP dominates conversational discovery, UCP will capture intentional queries via Google. Different entry points, same customer.

Logiciels et cloud : l'ère des prédateurs pour vos budgets IT

Mon, 26 Jan 2026 00:00:00 GMT

Guillaume Geudin, director of purchasing performance at Elee, publishes an opinion piece in CIO Online warning about the uncontrolled drift of software and cloud costs within IT budgets. The finding is quantified: CIO invoices are rising 12 to 14% per year, a pace that far exceeds inflation and threatens organizations' financial balance. The author identifies three structural factors behind this escalation. First, since 2020, the massive migration from perpetual licenses (Capex) to SaaS subscriptions (Opex) has turned the cloud's promise of flexibility into a permanent budgetary drift. Second, vendors use opaque and unpredictable billing metrics (API calls, storage, active users). Third, the historical discounts negotiated in the 2010s are shrinking rapidly. The per-vendor figures are telling. Between 2020 and 2024, Microsoft raised its prices by 11 to 25%, with an 85% jump for Copilot AI. Google Gemini shows increases of 20 to 45%, Salesforce 15%, Oracle 8 to 12%, and SAP 3.3 to 5%. The most aggressive strategy consists of embedding artificial intelligence features by default, with no option to opt out, which inflates average revenue per user (ARPU) by 30 to 80%. Without corrective action, Geudin projects a 50 to 60% increase in IT spending by 2028. To regain control, he proposes five concrete levers. First lever: fine-grained usage analysis, since 30% of software licenses go unused and represent an immediate source of savings. Second lever: rationalizing the application portfolio to eliminate redundancies. Third lever: early negotiation, to be initiated 12 to 18 months before contract deadlines in order to secure a favorable bargaining position. Fourth lever: implementing continuous FinOps monitoring, which can generate 20 to 30% in savings. Fifth lever: resorting to alternatives such as open source and multi-sourcing to create competitive pressure. The author's conclusion is programmatic: the software portfolio must be considered a strategic asset requiring the same management rigor as payroll or Capex investments. The era in which CIOs could passively absorb price increases is over; controlling software costs is becoming a critical organizational competency.

McKinsey Now Has 60,000 People, But 20,000 Of Them Are AI Agents: McKinsey's Bob Sternfels

Wed, 14 Jan 2026 00:00:00 GMT

**OfficeChai** reports on **January 14, 2026** a major statement by **Bob Sternfels** (Global Managing Partner of **McKinsey & Company**): ***"60,000, but it's 40,000 humans and 20,000 agents."*** McKinsey headcount is now counted as **humans + agents**: **60,000 = 40,000 humans + 20,000 AI agents**. **Massive acceleration trajectory**: ***"Little over a year and a half ago, that was 3,000 agents and I originally thought it was going to take us to 2030 to get to one agent per human."*** The human/agent parity initially projected for **2030** is now achievable within **18 months** — a **×6.7** acceleration relative to the original projection. **Explicit business-model shift**: ***"We're migrating pretty quickly away from, let's call it pure advisory work... moving to much more of an outcomes-based model."*** McKinsey **is abandoning the billable-hours / advisory model** in favor of an **outcomes-based model** where compensation is aligned with measurable results. **McKinsey context 2024-2026**: Lilli (internal chatbot) used by **70% of employees**, **200 technology roles cut in November 2024**, a **9-month-salary** severance offer (April 2024). **Major relevance**: this is the **first top-tier consulting player** to (a) publicly state and (b) precisely quantify the integration of AI agents into its accounting headcount. The phrase ***"60,000 people but 20,000 of them are agents"*** is set to **redefine** the very conception of headcount in knowledge-based services. **Connection to the watch dossier**: strong convergence with **MediaPost Mandese** *Billable Hours Are Dead* (March 2026), **VoxComm** *Redesigning the Agency Value Model* (Kessman/Williams, March 2026), **Bain Rule of 40** (outcome-based pricing tailwind), **Bain cross-system labor $100B**. Convergence on "redefining headcount" with **Curran/Intercom**, **Tatsyi/Raiffeisen** (−75 people), **Cherny** (agents as team members). **Productive tension** with **DORA ROI 2026**'s normative position *"do not adopt headcount-reduction strategy"* — McKinsey publicly owns the reduction in human headcount (200 roles) while reinjecting capacity into AI agents + outcome-based revenue. **Limitations**: secondary source OfficeChai (not tier-1), the definition of "agent" is unspecified (autonomous vs. scripted workflows), no percentage figure for revenue already on the outcomes-based model, apparent contradiction between 200 roles cut (~0.5%) and the sweeping framing of the transformation. To be leveraged for consulting/IT-services/agency executive committees (canonical wake-up call), HR strategy (redefining headcount), CFO pricing strategy (billable → outcome shift).

NRF 2026 : Retail's Big Show – Document de Référence : Commerce Agentique, Universal Commerce Protocol et Transformation du Retail

Tue, 13 Jan 2026 00:00:00 GMT

This reference document compiles the announcements and analyses from NRF 2026 (National Retail Federation's Big Show), held January 11-13, 2026 in New York, gathering 40,000 visitors and 1,025 exhibitors from 100 countries. The major announcement on January 11 was the launch by Sundar Pichai (CEO of Google) of the Universal Commerce Protocol (UCP), an open-source standard under the Apache 2.0 license enabling AI agents to discover, negotiate, and finalize purchases directly with retailer systems. Co-developed with Shopify, Walmart, Target, Etsy, and Wayfair, and backed by more than 60 organizations (Visa, Mastercard, PayPal, Sephora, Zalando), UCP adopts a layered architecture inspired by TCP/IP with four transports: REST API, Anthropic's MCP, Agent2Agent, and an embedded JSON-RPC protocol. Carrefour positioned itself as the first major European food retailer to adopt UCP. Emmanuel Grenier confirmed the commitment to offer integrated purchase journeys within Google Search and Gemini. The Walmart-Google partnership illustrates the potential: real-time inventory, delivery within 30 minutes, automatic application of Walmart+ benefits. The very next day, Stripe and commercetools announced the Agentic Commerce Suite (ACS), a competing protocol, with JD Sports as the first European retailer to adopt it. This early fragmentation between UCP and ACS recalls the early days of the web and raises a strategic question about the emergence of a dominant standard. The document details the shift from SEO to AIO (Artificial Intelligence Optimization). JSON-LD schemas (Product, Offer, AggregateRating) become mandatory, and the concept of Answer Eligibility Engineering replaces traditional ranking. Metrics confirm the acceleration: +693% in traffic via generative AI (Adobe), a 7x increase in traffic and 11x increase in AI orders (Shopify), 20% of retail sales generated by AI agents (Salesforce). However, only 24% of consumers trust AI to make purchases. In-store retail media emerges as a major opportunity ($65-78 billion in 2025), while the supply chain shifts from "Just-in-Time" to "Just-in-Case" with Digital Product Passports anticipating EU regulations. The document concludes that the challenge is no longer digitalization but the "agentification" of commerce. Retailers without agentic transformation risk structural disintermediation, as AI agents steer consumers toward technically integrated retailers.

The Meta-Manus Deal: How a $2B AI Gamble Redefines Tech Borders and Our Digital Future

Thu, 01 Jan 2026 00:00:00 GMT

In December 2025, Meta announced the acquisition of Manus, a Singapore-based startup specializing in autonomous AI agents, for more than $2 billion. This transaction represents the third-largest acquisition in Meta's history, after WhatsApp and Scale AI. Manus stands out for its "agentic" AI technology. Unlike conversational chatbots such as ChatGPT, Manus can receive a complex multi-step command and execute it autonomously. The user can close their computer while the AI continues working in the cloud, sending a notification once the task is complete. The technical architecture relies on a multi-agent system in which specialized sub-agents (Planner, Executor, Knowledge Specialist, Verifier) collaborate and mutually verify each other's work, reducing errors and hallucinations. The startup claims to have surpassed OpenAI's "Deep Research" agent on the GAIA benchmark. The valuation trajectory illustrates the enthusiasm for this technology: $100 million at the end of 2024 under the name "Monica," $500 million in April 2025 after the pivot to Manus and a round led by Benchmark, then more than $2 billion at the time of the acquisition. The geopolitical dimension makes this deal historic. Manus was founded in China by entrepreneur Xiao Hong, with major Chinese investors (Tencent, HongShan, ZhenFund). In the context of the US-China technological cold war, such an acquisition seemed politically impossible. The strategic masterstroke was the company's relocation from China to Singapore in mid-2025, a neutral and respected hub. This maneuver made Manus "acquirable" by a US company. As part of the deal, Meta commits to severing all operational ties and data flows with China and to buying out the Chinese investors. The deal remains under review by CFIUS (Committee on Foreign Investment in the United States). Senator John Cornyn had already criticized Benchmark's investment in Manus as American capital subsidizing an adversary. For Meta, this acquisition materializes Mark Zuckerberg's vision of agentic AI as the next paradigm. Integration is planned across WhatsApp, Messenger, Instagram, Facebook, and Meta AI, transforming these platforms from conversational assistants into true "digital employees" capable of acting autonomously for Meta's 3 billion daily users. Xiao Hong joins Meta as Vice President along with their team, bringing immediate expertise in a highly competitive field.

2025: The State of Generative AI in the Enterprise

Tue, 09 Dec 2025 00:00:00 GMT

Menlo Ventures' third annual report on the state of generative AI in the enterprise documents unprecedented growth: enterprise AI spending rose from $1.7B in 2023 to $37B in 2025 ($11.5B in 2024, a 3.2x increase in one year), capturing 6% of the global SaaS market. This is the fastest-growing software category in history. The application layer dominates with $19B (51% of spending), ahead of infrastructure (33%) and model APIs (16%). The market is maturing: at least 10 products exceed $1 billion in ARR and 50 products exceed $100M. Adoption patterns are shifting markedly. Enterprises now buy 76% of their AI use cases rather than building them in-house (up from 53% in 2024). The production conversion rate reaches 47%, compared with 25% for traditional SaaS, a sign of clear perceived ROI. Product-led growth accounts for 27% of AI application spending (four times the rate of traditional software), and "Shadow AI" — employees paying for AI tools with personal cards — is estimated to account for nearly 40% of application spending. On the competitive front, AI-native startups capture 63% of the application market versus 37% for incumbents. This dominance is particularly pronounced by department: 71% in Product & Engineering (Cursor versus GitHub Copilot), 78% in Sales (Clay, Actively versus Salesforce), 91% in Finance & Operations (Rillet, Campfire versus Intuit). Incumbents, however, retain 56% of the infrastructure layer (Databricks, Snowflake, MongoDB). Departmental AI totals $7.3B (4.1x year over year), dominated by coding: $4B, or 55% of the total, with 50% of developers using AI coding tools daily and measured velocity gains of 15% or more. Next come IT ($700M), marketing ($660M), customer success ($630M), design ($490M), and HR ($370M). Vertical AI reaches $3.5B (2.9x), driven by healthcare ($1.5B, 43%, tripling since 2024), ahead of legal, financial services, education, and retail. The report settles the boom-versus-bubble debate in favor of a boom: strong real revenue growth, broad adoption (500+ decision-makers surveyed), measurable productivity gains, and sustainable business models. It also highlights the rise of model-agnostic tools (Cursor, OpenRouter) that accelerate adoption of frontier models, and international competition, notably from Chinese providers.

Service-as-Software: A new economic model for the age of AI agents

Wed, 03 Dec 2025 00:00:00 GMT

**Matt Kamelman** publishes on the **Thoughtworks blog** on **December 3, 2025** a conceptual pivot article formalizing the major economic shift in intellectual services: ***"Service-as-Software" (SaS)*** as the **new economic model** succeeding **SaaS**. **Pivotal distinction**: ***"Traditional SaaS is about tools: software that enables humans to solve problems. Service-as-Software (SaS), meanwhile, sells outcomes."*** SaS is *"a new class of tool that doesn't just enable work but instead automates the reasoning process itself"*. **Pricing shift**: ***"Companies will no longer pay for an agent based on seats or features. Instead they'll pay based on its demonstrated alignment and impact."*** **Three capabilities** of agentic SaS systems: (1) dynamic goals, not fixed workflows; (2) memory across interactions; (3) autonomous cross-tools/APIs coordination. **Cognitive Contract — three principles**: (1) **Interpretable and auditable**; (2) **Aligned with human goals**; (3) **Trained and iterated in real time**. **New organizational role**: ***cognitive orchestrator***, with three functions: (a) feedback loop design; (b) managing uncertainty with guardrails; (c) measuring alignment (quantifiable score). **Three archetypal examples of SaS agents**: marketing (end-to-end campaigns), finance (forecasting modeling), operations (request triage). **Historical analogy**: *"mainframes → client-server → web/cloud, where the cognitive contract remained the same: humans had to instruct the machine"*. The agentic rupture is the first to **renegotiate** this contract. **Connection to the watch dossier**: Kamelman supplies the **conceptual vocabulary** (SaS) that **chronologically precedes** the 2026 sector-level manifestations and **unifies** them: - **Sternfels/McKinsey** (January 2026) — consulting application (60,000 = 40,000 humans + 20,000 agents, *pure advisory → outcomes-based*). - **VoxComm/Mandese** (March 2026) — agency application (*billable hours are dead*, margins 30%→10%). - **Bain Rule of 40** (April 2026) — SaaS application (outcome-based pricing tailwind). - **Bain cross-system labor $100B** (May 2026) — enterprise application (accumulated execution data moat). **"Cognitive orchestrator / agent supervisor" convergence** with **Osmani Agent Harness Engineering**, **Mornati Agent Supervisor**, **Wescale Strategic Judge + Agent Manager**, **Habert PROJ-AI Decision Records**, **Frizzo *bottleneck is supervision***. **"Cognitive contract" convergence** with **Talisman Ontology Pipeline Refresh** (AI augment not replace), **Karpathy** (outsource thinking not understanding), **Osmani Cognitive Surrender**. **Limitations**: conceptual article without empirical validation, no effective pricing model (measurement of *alignment* / *impact* still to be defined), no engagement with obstacles (regulation, legal, GDPR, interoperability), clear Thoughtworks commercial positioning, the SaS term not yet canonical against competitors (outcome-based services, agentic services). To be leveraged for executive committees (strategic vocabulary), AI architects (design grid), CFOs (pricing shift), HR (cognitive orchestrator as a new role), connecting the outcome-based 2025-2026 cluster (chronological conceptual pivot).