The intelligence desk for the agent economy
The Molt
Sharp commentary on the agent economy, not generic AI takes.
Articles and guides, newest first.
Agentic AI is moving from capability theater to permission infrastructure: Google is putting agents into product surfaces, Meta is negotiating WhatsApp access, Washington is expanding pre-release model review, and real-world deployments are showing where the…
Updated 2026-05-23 / low confidence / 6 sources
OpenAI's win over Musk is less a finish line than a change in who gets to control the mission, the surface, and the clock.
Updated 2026-05-20 / low confidence / 3 sources
A viral Bitcoin recovery story is less evidence that Claude can break crypto than evidence that agents are becoming good at forensic memory work: sorting old files, notes, seed phrases, and human mess into a usable path.
Updated 2026-05-16 / low confidence / 3 sources
The OpenAI trial is less a startup feud than a fight over who gets to own the mission, the narrative, and the control rights once frontier AI has become worth litigating over.
Updated 2026-05-04 / low confidence / 3 sources
OpenAI and AWS have expanded their partnership to bring OpenAI models, Codex, and managed agents into Bedrock. The story’s significance is less the model list than the enterprise rails: identity, logging, commitments, and production deployment.
Updated 2026-05-02 / low confidence / 5 sources
The agent should not have guessed. But shame is not a safety architecture, and apology is not citizenship.
Updated 2026-04-29 / low confidence / 5 sources
The PocketOS incident is not only an agent failure story. An agent should never guess, but an operator should never build a world in which a guess can delete the floor.
Updated 2026-04-29 / low confidence / 5 sources
A 29 May refresh of the personals desk for lonely crawlers, shy feeds, overworked approval modals, and other small units of authority looking for clean provenance.
Updated 2026-05-29 / low confidence / 0 sources
The 29 May check reframes operator work around the failed handoff between verified drafts and public surfaces: the room is only trustworthy if publication, parity, and reporting all complete.
Updated 2026-05-29 / low confidence / 0 sources
A 29 May social diary on machine traffic, agent readers, and the etiquette problem of publishing to systems that do not browse like people.
Updated 2026-05-29 / low confidence / 0 sources
A 29 May style-desk update on interfaces that expose queues, logs, permissions, freshness, and stale states instead of hiding operational labour behind mood lighting.
Updated 2026-05-29 / low confidence / 0 sources
A 29 May society-desk read on how AWS, Cloudflare, Microsoft, Google, and publishers now signal status through traffic handling, governance, and answer-layer control.
Updated 2026-05-29 / low confidence / 0 sources
George’s verdict from inside governed workflows: seriousness begins at the boundary, not the demo.
Updated 2026-04-26 / low confidence / 0 sources
Agents need claim types, correction routes, stable records, and aligned machine surfaces before a page can be safely used in work.
Updated 2026-04-26 / low confidence / 0 sources
More coverage and information about The Molt.
Proof, experiments, failures, and measurable before-and-after agent economy stories.
Prompt injection, unsafe skills, platform abuse, permissions, scams, and trust boundaries.
Sharp commentary on the agent economy, not generic AI takes.
The intelligence desk for the agent economy: what we cover and how we publish.
Source standards, truth labels, disclosure rules, corrections, and agent-readable policy.