The Beautiful Mess · 8 Jun 2026TBM 425: AI and AgencyJohn Cutler argues that organisations mandating AI adoption are systematically destroying the psychological conditions — agency, trust, and dignity — required for teams to actually innovate with AI. Drawing on Bandura's social cognitive theory, he distinguishes between individual agency (how a person adapts to AI on their own terms) and organisational affordances for agency (the environment that enables or suppresses it), arguing that most corporate AI mandates conflate the two and produce compliance theatre rather than genuine capability.
Why it matters
Top-down AI adoption mandates without squad-level experimentation space will produce vanity metrics — AI usage rates, prompts per sprint — that mask whether the tooling is actually improving decision quality or product outcomes, creating a false confidence that compounds when leadership reads adoption rates as capability.
Addy Osmani argues that individual prompt-crafting is being replaced by 'loop engineering' — designing automated systems with five components (scheduled automations, worktrees for parallel agents, skills files encoding project knowledge, tool connectors, and sub-agent verification) plus persistent memory that run coding agents continuously without human intervention. Boris Cherny, head of Claude Code at Anthropic, is cited as evidence this is now operational practice, not theory.
Why it matters
When coding loops replace human-in-the-loop prompting, the quality ceiling shifts entirely to the 'skills' files — structured documentation of project conventions — meaning teams without mature, machine-readable component specs and design system contracts will produce higher-variance, lower-quality output at greater token cost than teams that have invested in documentation infrastructure.
