How to Improve at Sensemaking AI
Why this is in the vault
Closes Cedric’s three-part sensemaking series, which is the most directly load-bearing Commoncog material for the MAC bet and the agent-deployer positioning. Part 1 said “ignore takes, read field reports of use.” Part 2 introduced Klein’s Data-Frame Theory of how experts sensemake under uncertainty. Part 3 fuses both into a method, grounded in the software-engineering AI controversy as the working example.
If MAC is going to teach operators how to deploy AI agents productively, the underlying theory of how operators improve their judgment about AI is part of the curriculum’s spine. Cedric is doing that theory work in public.
Core argument
The piece uses three competing programmer camps (the Karpathy/vibe-coding believers, the skeptics, and the disciplined-context middle) as a live case of frame divergence: same evidence, three different interpretive frames, all talking past each other. Cedric traces how the most controversial frame got constructed, then uses the example to teach two skills:
- Avoiding frame fixation (recognize when your frame is filtering out disconfirming data).
- Improving frame construction (build better frames by reading detailed field reports across multiple domains, not summaries or hot takes).
Read against Part 2’s Data-Frame Theory: experts sensemake by quickly toggling frames against incoming data and reframing when the data does not fit. Novices fix on the first frame and bend data to it. The path to expert sensemaking on AI is therefore deliberate frame-collection plus frame-stress-testing, not consuming more predictions.
Mapping against Ray Data Co
Strong relevance to MAC product design. MAC’s pitch is teaching operators to deploy agents well. Cedric is articulating, in the same period and with overlapping audience, the meta-skill of “how do I get smart about AI without getting captured by the discourse.” That meta-skill belongs in MAC’s foundation module - or at least gets cited there.
Concrete applications:
- Curriculum spine: the three-part series is a ready-made reading assignment for the “how to think about AI without becoming an AI bro or AI doomer” module. License-free linkable third-party scaffolding.
- Field-report discipline as MAC pedagogy: Cedric’s “read field reports, not predictions” maps directly to how MAC could structure case studies - real deployment narratives over abstract framework slides.
- Frame-fixation as the failure mode MAC prevents: many operators picking up Claude Code right now are stuck in one frame (it’s a toy / it’s magic / it replaces engineers). MAC’s value prop can be reframed as “we teach you the frame-toggling discipline so you don’t get stuck.”
- Sensemaking AI is itself a sellable wedge: separate from MAC, this could be a Sanity Check angle (the three-camp software controversy is concrete and recent enough to ground a piece).
Adjacent to the agent-deployer JD work (2026-04-14): the agent-deployer role description is light on judgment-criteria for the human in the loop. Cedric’s data-frame vocabulary gives that judgment criteria a name.
Sponsor disclosure
Self-CTAs to Commoncog Membership in two places (top and bottom). Standard for the newsletter, no third-party promotion. Member-forum thread roundup at the bottom is a soft member-acquisition tool but not paid placement.
Related
- 2026-04-19-commoncog-the-tacit-knowledge-series - foundation for why field reports beat takes
- 2026-04-19-commoncog-business-expertise-series - parent series on how expertise compounds
- 2026-04-19-commoncog-expertise-is-just-pattern-matching - Data-Frame is one mechanism for that pattern matching
- 2026-04-19-commoncog-tacit-skill-in-wicked-domains - AI deployment is wicked-domain par excellence
- 2026-04-19-commoncog-lia-dibello-business-expertise - DiBello’s mental-model work parallels Klein’s frames
- 2026-04-19-commoncog-breaking-out-sme-loop - what frame-fixation looks like in practice
- 2026-03-22-essay-architecture-human-shaped-sensemaking - vault’s prior cross-author take on sensemaking
- 2026-04-14-levie-agent-deployer-role-jd - the role this skill is for
- 2026-04-30-mac-bet-architecture-audit - MAC architecture this material plugs into
- 2026-04-20-indydevdan-agent-experts-self-improving - related: agents that build their own expertise