Pivotal essay by Dan Shipper (CEO Every) published on May 21, 2026 on every.to, "After Automation" — an argued response to the thesis of AI-driven collapse of knowledge work. Pivot thesis: AI progress creates more work for humans, not less.
By **Dan Shipper** — CEO et co-fondateur de **Every**// Source every.to ↗/Reading 2 min/.md// Auto-verified translation
#Dan Shipper#Every#after automation#AI commoditization cycle#commodification cycle#frame vs framer#the frame is not the framer#frame problem regenerates one level up
Summary
Dan Shipper, CEO of Every (AI-native media / studio), published a pivotal essay on every.to on May 21, 2026, titled "After Automation", an explicit counter-narrative to apocalyptic mass-unemployment narratives (Amodei, Sun, Wallace-Wells). Pivot thesis: "there's more work to do than ever" — AI progress creates more work for humans, not less.
Shipper formalizes the mechanism through a 5-step commodification cycle: (1) AI commoditizes yesterday's human skill; (2) that cheap skill is widely adopted; (3) abundance produces slop (sameness); (4) humans demand difference; (5) experts use AI to address today's problems, restarting the loop.
Central framework: the distinction frame vs framer. Benchmarks measure performance within specific frames — once saturated, changing the frame resets the counter. Models escalate within frames but do not replace framers. Pivot formula: "the frame is not the framer". Even at AGI, "the frame problem regenerates one level up" — a human directs the model toward a goal.
The "Human Sandwich": the human sets the frame upstream, AI executes, the human judges and extends downstream. Value shifts to both ends.
Two modes of working with agents: (a) agent employees (asynchronous delegation — Claudie, Andy, Viktor at Every; Fin at Intercom resolves 65% of support); (b) synchronous human-AI collaboration (Claude Code). At Every, 95% of CEO emails are handled by AI.
Benchmarks (May 2026): GPT-5.5 scores 62/100 on the Senior Engineer benchmark (human 80-90); GDPval measures 40-49% of expert human level, but requires extensive human framing. OpenClaw generated 44,469 PRs in May 2026 (vs Kubernetes' 5,200 PRs across all of 2022) — volumetric proof that agentic work produces more work.
Zeno's paradox of AI: Achilles (AI) runs toward the turtle (human), but the turtle "is alive to a specific moment", constantly moving toward new problems — Achilles never catches up.
Conclusion: this is not a tipping-point event, it is a persistent pattern that defines the future of work. Models optimize within the contexts humans specify; humans remain necessary to decide "what matters now". To be leveraged for COMEX: frame vs framer becomes the canonical 2026 grid.
Key takeaways
Date / source.May 21, 2026, every.to/p/after-automation, author Dan Shipper (CEO Every).
Pivot thesis.AI creates more work for humans, not less — an explicit counter-narrative to mass-unemployment narratives (Amodei, Sun, Wallace-Wells).
Position. operator of a 100% agentic company + theorist — a rare combination. ### The commodification cycle (5 steps) | Step | Description | |-------|-------------| | 1. Commoditize | AI makes yesterday's human expertise "cheap and widely available" | | 2. Adopt | The cheap skill is widely adopted → abundance | | 3. Slop | Abundance produces sameness — the slop | | 4. Demand difference | Humans demand difference → renewed demand for experts | | 5. Use AI | Experts use AI to address today's problems → back to step 1 | ### The Human Sandwich ` Human (frame the problem) ↓ AI (execute within the frame) ↓ Human (judge the output, extend toward next frame) `Strategic reading: human value shifts to both ends — framing upstream and judgment downstream. Intermediate execution skills are absorbed by AI. ### Frame vs. Framer — canonical concept > "The frame is not the framer." | Notion | Definition | |--------|-----------| | Frame | The specific framing of a problem — "senior engineer benchmark", "codebase rewrite", "GDPval task". Models improve within a fixed frame. | | Framer | The human who chooses the frame, defines what matters, interprets results, changes the frame when the previous one is saturated. | | Regeneration | "The frame problem regenerates one level up." — When AI solves a frame, the human framer sets a new one, above it. | ### Two modes of working with agents | Mode | Description | Every examples | |------|-------------|----------------| | Agent employees | Asynchronous delegation, persistent identity, operates in the background | Claudie, Andy, Viktor (named agent employees); Fin (Intercom) → 65% support conversations | | Human-AI collaboration | Synchronous tool, used like an augmented IDE | Claude Code and equivalents | → These two modes coexist, they are not mutually exclusive. This duality is what allows Every to have 95% of CEO emails automated and full-time human work. ### The Zeno's paradox of AI
Achilles (AI) runs toward the turtle (human).
Achilles closes 9/10 of the distance → the turtle has moved.
Achilles closes 9/10 of the new distance → the turtle has moved again.
Limit reached only at infinity — and only if the turtle stops moving.
*"Humans are alive to a specific moment". — the turtle never stops moving because it lives the running wants, running concerns of today. The model lives at its training cutoff. ### Benchmarks (May 2026) | Benchmark | Model score | Human score | Interpretation | |-----------|--------------|--------------|----------------| | Senior Engineer (codebase rewrite) | GPT-5.5: 62/100 | 80-90 | The model progresses but does not surpass humans | | GDPval (expert tasks) | 40-49% of expert level | 100% (ref) | + requires extensive human framing — so the metric itself is under-specified | | OpenClaw PRs (May 2026) | 44,469 PRs | (vs Kubernetes' 5,200 PRs across all of 2022) | PR volume ×9 — more work, not less | ### AGI implications > Even at AGI, "a human directs the model to achieve a goal". > The frame → execute → judge structure does not disappear — it moves up a level. Strategic consequence: preparing for "AGI that eliminates jobs" is mis-framed. The rational question is: "how do I structure my organization to exploit the regeneration of the frame problem?" ### Agents Without Agency (section title) The AI agent, even autonomous, has no agency — it has neither its own wants nor situated concerns. Practical consequence: a human is always needed to decide "what matters now"*. ### Watch-list linkage #### Strong convergence "redeployment toward framing, not replacement"
Shipper / Every. (2026-05-21): frame vs framer, commodification cycle.
Ng "No AI jobpocalypse". (2026-05-08): healthy US unemployment 4.3%, AI jobapalooza (job creation in AI engineering).
Tatsyi/Raiffeisen. (2026-05-05): "AI didn't make engineers faster, it made them different" — engineers redeployed toward "what engineers built that didn't exist before".
Mollick × roon ASI/FDE. (2026-05-10): "your own FDEs prove you know AI isn't self-adopting" — so human work remains structurally necessary.
→ Convergence: Shipper provides the theoretical framework (frame vs framer) that unifies the empirical observations of Ng, Tatsyi, Curran, and Mollick. #### Productive tension with apocalyptic narratives
Sun NYT "Permanent Underclass". (2026-04-30): Amodei white-collar bloodbath 50% by 2030, Block −50% Dorsey.
Wallace-Wells "AI Populism". (2026-05-08): Pew 50% worried, SF Molotov cocktail.
Mensch / Mistral. (2026-05-13): critique of Anthropic's "fear marketing".
→ Shipper provides the argumentative counter to these narratives: "the frame problem regenerates one level up" + quantified benchmarks (GPT-5.5 62/100, GDPval 40-49% expert). #### Convergence with harness engineering / context engineering doctrine
Shipper."the frame is not the framer" — the human sets the frame.
Osmani Agent Harness Engineering. (2026-04-19): "agent = model + harness" — the harness is the operational frame.
Seale Semantic Agent. (2026-04-17): "(Model+Harness) + (Ontology+Data) — ontology as the only moat" — ontology is the structural frame.
→ Convergence: frame = harness + ontology in agentic vocabulary; Shipper provides the business formulation of this technical structure. #### Convergence with Cherny Sequoia "coding is solved"
Cherny. (2026-05): "coding is solved", 150 PRs/day, 7 Powers reordering (switching costs ↓, process power ↓, network effects/scale economies/cornered resources unchanged).
Shipper."more work to do than ever" — consistent with Cherny since code itself is no longer the bottleneck (frame absorbed), but framing becomes the work.
→ Both say the same thing: human work shifts to the layers above. #### Tension with Osmani Cognitive Surrender
Osmani. (2026-05-05): Cognitive Surrender — "borrowing model's confidence as substitute for personal understanding".
Shipper."the frame is not the framer" — the human must remain an active framer, not surrender.
→ Coherence: Shipper structural + Osmani behavioral — "being a framer" requires "not cognitively surrendering". ### To be leveraged for
COMEX / boards.frame vs framer as the canonical governance grid for 2026 — "where are our people framers? where are they within the frame?".
HR / transformation. an anti-panic argument to explain to teams why AI increases work rather than eliminating it. Commodification cycle = vocabulary to frame redeployment, not layoffs.
CFO. a quantified counter-argument to "−50% headcount possible" narratives — GPT-5.5 62/100 + OpenClaw ×9 PRs as quantitative evidence.
Media / content CEOs. Every as a case study of an organization fully operated on agent employees + human-AI collaboration.
Product strategy.agent employees vs human-AI collaboration as a decision grid for which type of agent to build.
2026 strategic vocabulary.frame vs framer, commodification cycle, human sandwich, Zeno's paradox of AI — terms to include in COMEX presentations, executive training.
The knowledge graph extracted from this fiche — 13 entities, 22 relations.
In this graph :Dan Shipper · Every · After Automation · Commodification cycle · Frame vs Framer · Human Sandwich · Zeno's paradox of AI · Agent employees · Human-AI collaboration · Senior Engineer benchmark · GDPval · OpenClaw · Alive to a specific moment