Behavior and attitude, the new Agile approach - Nigel Thurlow (Talk Outline)

Nigel Thurlow — former Chief of Agile at Toyota, creator of Scrum the Toyota Way, co-creator of The Flow System, CEO of the Flow Consortium — English, lives in Texas (hat and boots, per a promise to his wife). A deliberately fast, funny, and alarming talk: we don’t have a methodology problem, we have a behavior and leadership problem — plus a rapid-fire AI tour.

1. The State of Agile (2025)

• Riffs on his “Jobs” talk: no more certification/PDF factories, made-up roles (“no such thing as a flow master”), and the Cynefin “big picture” now behind a paywall (“a secret society”).
• The transformation myth: the Agile Manifesto “beautified into sainthood” to prescribe how to transform a company — ”it’s just a philosophy.” And the misuse of culture — ”culture is a product of our behaviors.” (Nearly rewrote the talk after Kent Beck’s keynote to focus on his distributed leadership / psychological flow / team science work.)

2. Leadership Issues

• We don’t need more methods/tools/techniques — we need better behavior, starting with leadership.
• The reality of change: if you earn $1M+/year (some execs $45M), you don’t have a problem — the system you built works, so without significant fear/pain, nobody changes.
• America runs on incentives — bonus, status, power, income — with short horizons (OKRs, one to three months) and metric-worship: the McNamara quantification fallacy (focus on the metrics, ignore contradicting information — coined by an ex-Ford president).
• Leadership isn’t a job title — you earn it by building followers (followership). The test: “if you stopped paying them, would they still follow you?” Lean/agile fails because orgs desire change but aren’t designed for it, and leaders “are all right” and make every excuse not to change.

3. The AI Threat (fast and alarming)

• Timeline: transistor (1947) → AI (2017) → ChatGPT (2022) → agentic AI in 2 years. Yoshua Bengio (neural-network Nobel laureate): AI’s task capability doubles every ~7 months; “~5 years from a lot of trouble.” Geoffrey Hinton: “a very special point in history where everything might totally change at unprecedented scale.”
• Party tricks that matter (he’s 62, ex-COBOL/JCL/punch-tape, “can’t code anymore”): a Christian-rock Disney song built in 90 minutes (Suno for music/vocals from adapted Lion King lyrics, LTX Studio for storyboard/imagery); a Grok self-portrait + HeyGen persona video; and Gemini writing ~800 error-free lines in 2 minutes from a “stupid little prompt.” “It’s not what it wrote — it’s the fact it wrote it.” Yes it hallucinates (“put glue on pizza”), “but plenty of humans write bad code and we don’t remove them” — and it’ll get better faster than us.
• Agile coaches are “the dodo” — extinct unless they evolve. “Who is my customer? They don’t want agile coaches anymore.” The “field of dreams” fallacy. What leaders actually want lean to be: cut headcount, deliver faster, more profit, more status (Goldratt “pandered to their greed — it’s all about money”).
• Physical AI = robotics + AI + sensors/actuators/manipulators + world simulation:
  • NVIDIA Omniverse/Cosmos + Accenture + KION — warehouse digital twins (Mega blueprint; robots perceive/reason/plan/act in an Omniverse twin; simulate infinite scenarios before deploying). 20M+ hours of simulated video free on GitHub; chips as low as $350; Hugging Face.
  • Pegatron (Fortune 500) digital-twin case study; but ”AI won’t solve bad leadership” (Boeing and a collapsing Japanese carmaker use digital twins too).
  • Wetnee Robotics laundry-folding robots; Waymo in Austin (20% of rides, “no tip, doesn’t text and drive, clean, doesn’t drive like an idiot”); Tesla Optimus (“Musk’s next product, not cars — China will eat him for lunch”; they rode in a Chinese-built MG EV).
  • A self-learning LLM with zero input data (self-trains — “scary as hell”); Google AlphaEvolve writing/testing/debugging its own code and iterating — “AI doing agile all by itself.”
• Jidoka (Sakichi Toyoda, ~100 years ago): ”machines with a human-like intelligence… the human should never be removed; machines should serve humans” (“there is a beauty in the human touch”). Agentic AI is probabilistic, not deterministic — perceive/reason/plan/act autonomously.

4. What to Do

• Change behavior, not mindset (John Shook, lean: “we don’t change how people think, we change how people behave — we give them alternatives”). Stop the “seductive allure of neuroscience” (brain pictures make people believe anything) and “mindset” talk (pseudoscience — an individual has a mindset; you can’t collectively share one). Appeal to leaders’ actual desires (P&L) — why the big consulting houses thrive.
• EAST framework to “sell to leaders”: make it Easy (desired = default), Attractive (design incentives in), Social (reciprocity), Timely (stress immediate costs/benefits).
• New coaching roles (beyond “353 of Scrum”): ethical-risk coach, cyber-risk coach (though “AI auditing AI” already exists), and project-turnaround specialist — pull the metaphorical andon cord: stop/slow completely, stabilize (remove variation), address tech debt (“not a problem — a bad decision, usually by management”), eliminate red tape/non-value-add, make quality central, run weekly stakeholder reviews. Coaches evolve into orchestrators/conductors bridging worlds.
• Agentic = giving agency = power of attorney — you outsource the decision; you’re no longer the owner. Dark examples: Anduril’s Roadrunner autonomous weapons (“smart weapons vs. dumb weapons”), a Chinese factory robot beating its handler, and human enfeeblement (we forget how to change a tyre, read a map).

5. Q&A

• Tasks AI can’t replace? Trades — plumbers, electricians, mechanics (until cars go autonomous/electric). Diversify your skills portfolio; beware human enfeeblement. “To become valuable, learn skills beyond coding.”
• Keeping AI in check with its own output? He doesn’t fully know — it’s an arms race (“information is the new oil”), unfettered, with LLM API hacks/threats already happening → must be governed at governmental/legal level; we must become experts at controlling it.
• Organizational Behavior Management / behavior change + balancing AI? Give people opportunity to exploit their knowledge/skills, allow mistakes and support failure — the andon cord = psychological safety to stop work; “where’s the andon cord in software when tech debt piles up?” The lean view: AI is augmentation/assistant/tool — the danger is greed (billionaires now crave power, not money).
• Which of the seven AI levels are we at? No idea — can’t keep pace; most worried about coding/development ability; “we’re very close to AGI.” Start studying what it means for your life/family/community, not just tech, and reskill.
• When to say “no more AI” in a project? “No answer — 8 seconds ago your boss already decided; you won’t have that decision, and that’s the sad part.”

People & References Cited

• Nigel Thurlow — speaker; ex-Chief of Agile, Toyota; The Flow System, Flow Consortium.
• Yoshua Bengio, Geoffrey Hinton (AI/neural nets), Sakichi Toyoda (jidoka), John Shook (lean/behavior), Eli Goldratt (Theory of Constraints, criticized), Palmer Luckey / Anduril, Robert McNamara (fallacy), Mike Rother (Toyota Kata), Kent Beck / Dave Snowden / Dan North / Diana / Daniel Vacanti / Joseph / Henrik (co-speakers referenced).
• Tools: Suno, LTX Studio, Grok, HeyGen, Gemini, NVIDIA Omniverse/Cosmos, AlphaEvolve, Waymo, Tesla Optimus, Hugging Face.
• Concepts: culture-as-behavior, followership, McNamara fallacy, jidoka, EAST, andon cord, human enfeeblement, agentic = power of attorney, change behavior not mindset.

Video: https://www.youtube.com/watch?v=KBufxUOF7yA — Transcript via yt-transcript.sh; outline generated from the transcript.