note 2026-07-07 addition — the pyramid→diamond org-side view Jeff Crume's pyramid→diamond model (AI Gave You A Promotion (IBM Technology)) is the role-geometry twin of Ryberg's tasks→responsibilities shift. Where Ryberg names what changes at the individual scale (unit of work moves from task → responsibility), Crume names what changes at the org scale (the shape moves from pyramid → diamond as former entry-level people rise into an enlarged experienced-tier middle). Same underlying transformation, two different lenses; both anchored on Jevons Paradox as the scope-elasticity mechanism.

Tasks to Responsibilities Shift

Felix Ryberg's framing (June 2026, with Fable 5 launch): there have now been three eras of how people use AI.

Era	Unit	Human-in-loop
1. Chat	Questions — smarter search / autocomplete	Every turn
2. Tasks	A bounded problem handed to the model	Start and end
3. Responsibilities / loops	Ownership of an outcome over time	Set scope; don't sit in the loop

The canonical Ryberg example: "I no longer tell Claude to investigate a particular crash report. It runs a loop watching every crash report that comes in. Its job is no longer to help me fix a crash. It's to keep our apps from crashing."

Why it's structurally different

No start/end. A task is queued and dequeued; a responsibility is resident. The agent is always-on against a continuous stream.
Outcome-owned, not output-owned. The unit isn't output (a PR, a fix) but a bound on undesirable states (no crashes; no slow pages; no unread Daily Brief tickets).
Implies infrastructure. Continuous loops need triggers, observability, escalation, and budget controls — which is why this episode's adjacent conversations are all about Token Maxing and Uber's $1,500/employee/month cap.

The same idea from different angles

Boris Cherny (last week, CNBC): "I have loops that are running. They're the ones that are prompting Claude. My job is to write loops."
Peter Steinberger (X, the same week): "You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents."
Nate B. Jones / Task Imagination: the user-side framing — your bottleneck is imagining what to set as a responsibility.

Together these converge on Loops as Core Primitive as the engineering form.

Cross-references

Loops as Core Primitive — the implementation shape.
Task Imagination — the prerequisite literacy.
Auto Research Loop (Karpathy) — a specific kind of loop (research/optimization); responsibility-class is broader.
How Loops Are Improving Work — Sunil's Research Brief — the vault's existing taxonomy. The responsibilities framing slots cleanly above scheduled / OODA loops; consider re-cutting the brief with this as the new spine.
Bike Method — Nate Herk's phased-autonomy frame. Phase 4 of Bike Method ≈ responsibility-level operation; phases 1–3 are still in the tasks era.

Connects to your work

process-raw is already a responsibility, not a task — it watches the raw/ folder, makes ingest decisions, commits. That's the exact pattern Ryberg names. Worth saying out loud in any internal P&G IT discussion: we already run a small responsibility; here's what shipping more of them looks like.

2026-06-13 — The engineer-side version of the shift

The 7 Skills You Need to Build AI Agents (IBM Technology) (Bri Kopecki) names the same move from the builder's seat: engineers stop writing prompts (a task — craft a sentence, get an output) and start owning the reliability, security, and evaluation of autonomous agentic systems (a responsibility — keep the agent from failing, leaking, or regressing, continuously). Her line "the prompt engineer got us here; the agent engineer will take us forward" is the tasks→responsibilities transition expressed as a role change.

2026-06-21 — The C-suite version (Zehren / Info-Tech LIVE 2026)

How Agentic AI Is Changing the CIO Role (Forbes, Tim Keary) names the same shift one altitude up, for the CIO: traditionally, CIOs were tasked with maintaining IT infrastructure (task era — bounded function, queued and dequeued) → now the bar is to drive innovation and connect technology to real business outcomes (responsibility era — resident, outcome-owned, no start/end). Tom Zehren's coinage Exponential IT is the executive-tier name for the same role mutation Ryberg names at the engineering layer. The labour-market evidence the C-suite version of the shift is happening, not just being talked about: CIO Transition Wave.

2026-06-27 — The macro-empirical mirror: the AI Productivity Disconnect

Does AI Adoption Improve Productivity (BOK Issue Note 2026-12) is the negative-space version of this shift in representative national survey data. Across 5,512 Korean workers, AI saves 3.8% of work time — but the correlation between worker-level time savings and worker-level output growth is zero. The exception groups in the regression — self-employed, professionals, top-50% AI usage — are exactly the workers already operating in Era 3 (resident outcome ownership, performance-tied compensation, high autonomy). The disconnect is the population still operating in Era 2.

In other words: Ryberg's "tasks → responsibilities" is not just an engineering insight, it's the mechanism that closes the productivity disconnect. Until the bulk of work is recut from task-list-execution to responsibility-ownership (with workflow / incentive / verification redesigned around it), the BOK data predicts the macro productivity gain stays near zero. The Solow Paradox is the historical antecedent — same shape played out for electricity and PCs.

2026-06-20 — The market-priced version: TCS / Indian IT outsourcers

Tatas Big Bets Are Yet to Pay Off (Economist) is the market-implied version of the shift hitting an entire industry. Tata Consultancy Services lost >50% of market cap since end-2024 even while operating profit rose +12% — investors are pricing the AI-undermines-Indian-IT-outsourcers thesis ahead of any earnings impact. The Indian-IT outsourcing business is structurally a "tasks-era" service: bounded, headcount-priced, queued-and-dequeued. If the world recuts work from tasks to responsibilities, the labour-arbitrage code-factory model loses its anchor.

The Economist's framing: "Tata Sons must find a strategy for India's largest IT firm to adapt to AI." That sentence is exactly the tasks-to-responsibilities transition expressed as a corporate-strategy question for India's largest tech employer. See Indian IT and AI for the full concept.

2026-06-27 — Anthropic now separates the surfaces that embody Eras 2 and 3 in its own data

Anthropic Economic Index Cadences Report (June 2026) is the first time Anthropic publishes usage data with the surfaces themselves split out — and the split tracks Ryberg's eras:

Claude Cowork = the Era-2 tasks surface: multi-step bounded execution, human at start and end. Cowork+chat artifact mix is dominated by explanations and guidance.
Claude Code = the Era-3 responsibilities surface: resident loops, agent watches the stream (Boris's 150 PRs/day, the "loop watching every crash report" pattern). Claude Code's artifact mix is far heavier on code and technical work, with different cadence patterns (weekend gaming and quant-trading clusters rise; backend architecture clusters fall).

The data-side framing closes the loop on what was a Ryberg-coined product-marketing claim. Cowork sits in the tasks band; Code is operationally in the responsibilities band — and Anthropic's analytics now treat them as separate cohorts.

2026-06-27 — The executive-tier consequence: hyperconvergence + the hourglass

A Leaders Guide to Advanced Team Structures (AWS Events) (Brovich) is the executive-tier team-and-org-design consequence of the same shift. Brovich's named "hyperconvergence": the new team is 2–3 expert generalists + agents, each owning a workflow end-to-end — i.e. a responsibility, not a lane. Coordination overhead and handoffs collapse. The pod is the Era-3 team shape; the Hourglass Organization is the organisation that houses pods and preserves the apprenticeship rung that produces the next generation of expert generalists. The Era 1 → 2 → 3 progression on this page describes the unit of work; Brovich's pyramid → diamond → inverted-pyramid → hourglass progression describes the unit of team that maps onto it.

2026-07-03 — The shift as a hiring criterion (Ramp/Box)

AI Companies Are Hiring More (AI Daily Brief) shows the shift showing up in who firms hire. Ramp's economist: high-AI-adoption firms are "hiring different kinds of employees… people who know how to use AI and use it well." That is the tasks→responsibilities move expressed as a recruiting filter — hire people who can own AI-run loops, not just execute bounded tasks. Aaron Levie's expansion mechanism (AI grows scope → more customers / more software → hire more people) is the demand-side complement: when the unit of work becomes an owned outcome, capable owners become the binding constraint, so adopters pull more of them in — especially at the entry level, where AI-native habits can be trained from day one. See AI Adoption and Headcount Growth.

Sources

Fable 5 Raises the Bar for AI Ambition — Ryberg's canonical framing
The Way We Use AI is Changing — Cherny + Steinberger reinforcement
How Agentic AI Is Changing the CIO Role (Forbes, Tim Keary) — the executive-tier translation
Does AI Adoption Improve Productivity (BOK Issue Note 2026-12) — the macro-empirical negative-space of the shift
Tatas Big Bets Are Yet to Pay Off (Economist) — 2026-06-20; the market-priced industry-level case (TCS)
Anthropic Economic Index Cadences Report (June 2026) — Anthropic now splits Cowork (tasks-era) from Code (responsibilities-era) in its own usage data
A Leaders Guide to Advanced Team Structures (AWS Events) — Brovich's "hyperconvergence" team shape + hourglass org that map onto the Era 1→2→3 progression
AI Companies Are Hiring More (AI Daily Brief) — 2026-07-02; the shift as a hiring criterion ("hire people who use AI well") + Levie's scope-expansion mechanism
YC Summer 2026 Requests for Startups (The VC Corner) — 2026-07-08; the firm-level extension of the shift. YC's Summer 2026 RFS #2 AI-Native Service Companies (Alströmer) explicitly promotes Ryberg's frame from individual scope to business-model scope: an AI-native services firm holds the responsibility (a filed return, an approved claim, a closed compliance case); the agents inside execute the tasks. Six of YC's 15 RFS items encode the same "human sets scope, agents run resident loops" pattern (Company Brain, AI OS for companies, dynamic UIs, software for agents, SaaS challengers, AI-native services). Reads as market-signal confirmation that tier-1 US VC is now sizing companies against Era-3 unit economics.

2026-07-21 — The workflow altitude (Uber's Agentic Pods)

How to Help People Thrive with AI (AI Daily Brief) introduces a fifth altitude of the shift, sitting between team (Brovich) and firm (Alströmer): the cross-functional workflow itself. Praveen Napali (Uber CTO): "The workflow becomes the unit of automation, not the individual task. The most impactful agent skills cut across teams, orgs, functions, tools, and systems."

Uber's Agentic Pods programme is the operational cadence for this altitude: 30 engineers × 30 domain experts × 2-week timebox × 16 functions × 2 months = a cross-functional workflow catalogue where the workflow holds the responsibility and the agents inside it run the tasks.

Updated altitude ladder:

Individual (Ryberg, 2026-06): unit of work = responsibility, not task
Team (Brovich, 2026-06): 2–3 expert generalists + agents = hyperconvergence pod, owning workflow end-to-end
Workflow (Uber / Napali, 2026-07): cross-functional workflow is the unit of automation; the pod is the temporary structure that discovers and encodes it
Org (Zehren, 2026-06): CIO role recut around business outcomes, not IT infrastructure
Firm (Alströmer/YC, 2026-04): business model = "sell the outcome, not the tools" — AI-native services

The Uber datapoint also validates Brovich's "someone using AI will take your job" reframe from the other direction: in the pods programme, the domain expert is co-building the agent that runs their own workflow — they're not being replaced by it, they're becoming its owner.

2026-07-08 — The shift is now a firm-level business model (YC S26 RFS)

YC Summer 2026 Requests for Startups (The VC Corner) is the vault's first source in which the responsibilities-era frame appears as a business-model design, not an individual-worker or team-shape one. Gustaf Alströmer's AI-Native Service Companies item ("2023 to 2025 was the era of AI copilots. The next era is companies that skip the human entirely and just do the work") is Ryberg's tasks→responsibilities shift promoted one altitude — from the individual owns the responsibility, the agent runs the tasks to the firm as a whole owns the service outcome, the agents inside run the tasks. The vault has now seen the frame at four altitudes:

Individual (Ryberg, 2026-06): unit of work = responsibility, not task
Team (Brovich, 2026-06): 2–3 expert generalists + agents = hyperconvergence pod, owning workflow end-to-end
Org (Zehren, 2026-06): CIO role recut around business outcomes, not IT infrastructure
Firm (Alströmer/YC, 2026-04): business model = "sell the outcome, not the tools" — AI-native services

Each successive altitude widens the aperture; the underlying transformation is the same. YC's willingness to fund an entire batch shape around this at the firm level is second-source confirmation independent of the individual/team/org observations.