Twenty years of agile coaching failed to fix the micromanager who meddles with every draft, every meeting, every decision. This article shows where their distrust stops damaging teams and starts producing the verification work AI adoption actually needs. Welcome the Verification Architect!
Welcome to the 544th edition of the Food for Agile Thought newsletter, shared with 35,582 peers. This week, Taylor Pearson locates the real leverage of AI knowledge work tools in context-rich scaffolding that encodes local knowledge, a thesis Teresa Torres demonstrates in practice by fixing AI-generated Opportunity Solution Trees through agentic validation loops. Simon Willison watches that same agent reliability erode his own code-review discipline, and Bedard et al. name the resulting cognitive cost “AI brain fry.” Stephanie Leue and Len Greski shift the lens from individuals to systems: she with her 40/40/20 alignment rule, he with 90-day outcome-tied funding cycles.
Next, Aakash Gupta and Pawel Huryn argue that PMs should build a self-improving AI operating system instead of treating Claude Chat as the main interface, an investment in compounding leverage that Jeff Gothelf warns can backfire when AI-generated MVPs outpace the team’s ability actually to learn from customers. Allan Kelly looks to Ukraine for evidence that mission command and motivated teams beat rigid planning under real constraints, while Mike Fisher applies Grant’s wolf-counting lesson to deflate SaaS doomsday narratives. Also, the DORA team grounds the broader AI ROI debate in organizational maturity rather than tool choice.
Lastly, Kyle Poyar shows practitioners how to package 15 years of GTM expertise into reusable Claude skill files for research, pricing, and ICP work, the kind of individual leverage Robert Glaser warns rarely scales into organizational learning without his proposed “Loop Intelligence.” Ara Kharazian reports that Anthropic has overtaken OpenAI in business adoption at 34.4%, though cost and reliability cloud the lead. Finally, Grant Harvey finds organizations, not workers, are the real AI bottleneck, while James Shore models how unchecked coding-agent output quietly doubles maintenance debt.
You configured Claude in Claude Desktop, wrote instructions, uploaded reference files, and set your preferences. Then you clicked the Cowork tab.
Unfortunately, Claude had no memory of what you just did. Your instructions were gone, as were your files and preferences. You assumed this was a bug, but it is a feature: You switched applications.
Welcome to the 543rd edition of the Food for Agile Thought newsletter, shared with 35,597 peers. This week, John Cutler warns in his AI Playbook that AI accelerates broken practices just as much as good ones, while Guy Champniss adds that the psychological toll of AI adoption on employee motivation may quietly erase the productivity gains organizations expect. Teresa Torres and Petra Wille challenge the rising “taste” narrative in product management, proposing that discovery skills and evidence beat gut feeling. David Pereira diagnoses why product management in Europe often underperforms, pointing to roadmap theater and consensus paralysis, and Laura Klein reminds us that involving engineers in research builds shared understanding faster than any handoff ever could.
Next, Andrej Karpathy describes how software development is shifting from vibe coding to agentic engineering, where judgment and architectural understanding matter more than writing code, and Cedric Chin offers three techniques for making sense of AI without losing your current frame. Also, Richard Kasperowski reminds us that CEOs never cared about Scrum, only results, and that AI raises the bar on engineering fundamentals. Paweł Huryn shows how Claude Design compresses product discovery from weeks to hours, and Wyndo and Dheeraj Sharma walk through building a competitor intelligence agent that develops editorial judgment over time.
Lastly, Jake Handy warns that executives mistake AI token consumption for productivity, while Paul LaPosta shows how AI-generated code inflates DORA metrics by boosting speed while eroding system understanding. Jack Clark raises the stakes further, estimating a 60%+ chance of fully automated AI R&D by 2028. Jim Lewis, Jeff Sauro, and colleagues find that AI usability evaluations catch only half the problems humans identify while generating unverified issues on top. Finally, Mike Fisher reminds us that belonging is not a soft perk but a measurable performance driver, built or destroyed one manager interaction at a time.
If you have spent the last twenty years arguing that velocity is not value, that adoption is not impact, that an Agile transformation is not a Jira migration, the Stanford AI Index 2026 will read like déjà vu: The technology is new. The failure mode, the AI spending trap, is not.
The many organizations that have adopted AI but cannot show an EBIT impact are the same organizations that adopted Scrum without learning empiricism, adopted DevOps without changing how they fund teams, and adopted product management without giving anyone product authority.
The economic data is the evidence. The interpretation is what you, the agile practitioner, already know.
Welcome to the 542nd edition of the Food for Agile Thought newsletter, shared with 35,608 peers. This week, Mario Zechner advocates for slowing down with AI, warning that unsupervised coding agents compound errors faster than teams can fix them. Stephanie Leue shows how AI-driven speed tempts teams to skip discovery, incurring a hidden “Alignment Tax,” while Jenny Wanger and Michael Goitein find lasting advantage in internal capabilities, not copyable features. Mark Nottingham flags AI agents bypassing browser-level protections, Wharton’s Blueprint examines barriers to adoption, and Joost Minnaar uses the Titanic to show how silos filter critical signals.
Next, Teresa Torres and Petra Wille challenge the reflex to centralize decisions when uncertainty hits, arguing that real leadership sets direction and builds trust. Pawel Brodzinski extends that theme to AI-generated specs that look complete yet erode the human communication that teams need. Maxim Massenkoff shares Anthropic’s survey of 81,000 users, revealing that early-career workers worry most about displacement. Matthew Littlehale recounts replacing Scrum with Shape Up, and Andrej Karpathy reframes LLMs as a new computing paradigm requiring human judgment throughout.
Lastly, Steven J. Vaughan-Nichols warns that enterprise AI lock-in runs deeper than executives admit, with failed migrations and rising costs compounding quickly, while Kevin Kelly frames this instability as part of a broader “Age of Ambiguity” that demands radical adaptability. Dave Snowden surfaces a foundational tension within the CRP tradition between facilitated practice and radical process ontology. On the practical side, Michael Crist guides non-technical professionals through setting up Claude Cowork, while OpenAI’s GPT-5.5 prompt guidance shifts toward outcome-first instructions over process-heavy prompts.
