Evelyn Van Kelle – Podcast Stage | Craft 2026 (Conversation Outline)

A podcast-stage conversation at Craft 2026 (last of the day, sitting in the Budapest sun), recorded by the ABK Podcast / Leadership Anonymous (~4 years, a Slack community of ~250 mostly-leaders, founded to improve leadership in Hungary — “anonymous” so it’s about how to do better next time, not blaming past bosses). Moderator: Fanni / Karolina Tóth (cognitive scientist → IT consultant → high-performance coach → scuba dive leader). Guest: Evelyn van Kelle — social scientist and change-management consultant working at the intersection of social science and software systems, a self-described fan of heuristics. Thesis: the AI question isn’t “will it replace human judgment?” — it’s how do we, as humans, relate to this new technology and balance the social with the technical? Topics: human judgment, heuristics vs. frameworks, constructive disagreement, reinforcement, AI bias/authority, and leadership.

1. Framing: The System We Neglect Is Human-to-Human

• The industry obsesses over distributed / legacy / complex systems but neglects the human-to-human system — how we talk, the culture, how we use tools, how we decide — and it’s all tangled up.
• We prefer to phrase problems as tool problems (“one more feature and I won’t have to have the hard conversation”) because tools feel easy to fix.

2. Will AI Replace Human Judgment?

• Evelyn is “triggered” by the question — “let’s hope it never does, and I don’t think it ever will.” It’s not an either/or.
• Human judgment is contextual, situational, socially embedded — interpreting social dynamics, cognitive biases, handling conflict, reading a room, emotions. “AI will never do that the way we do; if it really replaced that, it’s not a world I’d want to work in.”
• The hidden assumption in the question — that the technology is the biggest challenge — is wrong; the tech part is not more important than the social/human part. The real question: how do we balance them, how does one impact the other, and how do we relate to it?

2.1 On layoffs

• Firing people to pay for AI is ”short-sighted”; companies will regret it.
• What we need more of is people who can question their own reasoning — “AI can’t do that for us.”
• Should everyone become a “tech person”? No — the tech people we already are should get more literate in the human sciences, because “the technology will be there; how we relate to it is what matters.”
• Fanni: it was just in the news that some CFOs realized the people they fired made more money than the AI tokens — “human capital is a capital.”

3. Heuristics vs. Frameworks

3.1 The distinction

• A framework is a structure you apply — often prescriptive (e.g., SAFe — “I’m triggering some people”).
• A heuristic is a rule of thumb developed personally from experience — you know what works because you’ve been in the situation.

3.2 Personal examples

• When she feels overwhelmed/anxious/nervous → go outside (even 5 minutes) — a heuristic learned from experience.
• In collaborative modeling sessions, when she feels “we’re stuck / people aren’t happy with me” → take a few steps back and observe the room: is this a me problem or a group problem?

3.3 The AI problem with heuristics

• Everyone has valuable personal heuristics, but they’re rarely documented; frameworks are well documented.
• So AI output is more framework than heuristic (based on what was written down), meaning we risk ignoring/neglecting undocumented heuristics.
• The tension: an experienced person’s gut feeling (“this solution won’t work”) vs. AI’s confident “this will definitely work” — do you fully trust the AI framework, or honor the experience-based heuristic? We need the heuristics.

3.4 Keep a heuristic journal

• She “preaches” a heuristic journal — write down “this works in this situation” across your different contexts (personal/work). You’re often unaware of your heuristics because they’re System 1 / trained.
• Heuristics develop over time (not fixed like a framework’s V2), which makes them more valuable.

3.5 Fanni’s story — “I’ve done this before”

• After her father died (a “last living relative”; he clung to possessions because Hungarian communists had taken his aristocratic family’s things), she inherited a house full of stuff and felt she’d “never get rid of it all.”
• She realized she’d done it before — at 17 she cleaned out her late grandmother’s apartment in Debrecen (3 hours away, no driver’s license), bringing friends by train and making it fun (once forgetting the keys in Budapest).
• “I’ve done this before, so I can do it again” — it took 8 years, but she sold the emptied house last year. Even undocumented, the past-self memory gave her trust in herself to go easier on herself.
• Evelyn: that’s exactly what heuristics should do — give confidence/trust (“I’ve been here before, I know what works”), and we should rely on them more in the AI world.

4. We Are Not Rational — and Neither Is AI

4.1 People rarely consult themselves

• People pretend to be rational, bring numbers to justify a conclusion, and call it a day; inner guidance is rarely consulted, and AI is assumed to “know” because it has “all the written notes ever.”
• Evelyn’s two talks covered cognitive bias in decision-making — “we are not rational human beings and never will be,” so we should question our own reasoning way more (reflect, take a step back: did I miss something? are there alternative perspectives?).

4.2 The value of constructive disagreement

• Learning and getting better lives in the uncomfortableness of not agreeing.
• AI is often treated as an authority (confident, well-formulated writing that fits your line of thought), so we’re tempted to go with it — but then we miss the uncomfortableness of disagreement and “a lot of wisdom and learning.” That’s a huge risk.

4.3 Who’s responsible when the AI is wrong?

• A fair but under-discussed question — most companies haven’t made agreements about ownership; even where agreements exist, follow-up (“okay, things went wrong, now who’s responsible?”) is “up for discussion again.”
• Her advice: don’t just assign one responsible person and don’t rely on a matrix/framework (just “a thing on paper”). You must go through the uncertainty/uncomfortableness, make mistakes, trial and error, then have the conversation about what works in this specific socio-technical context (every context is different).
• What’s needed: a culture where people feel safe to raise the questions (“hey, we haven’t discussed this”). But today such conversations are usually about blame (“who went wrong so we can feel better”). Culture takes time — “we need to make that time; it won’t happen overnight.”

5. “You Get What You Reinforce”

5.1 When is it okay to disagree with the AI?

• A cultural and behavioral-design question: you get what you reinforce.
• As a team/company, start by asking what do we value? If you say “we value being able to override AI suggestions,” the follow-up is: are you actually reinforcing that behavior?
• When someone overrides an AI suggestion, how do you respond? Reinforce it (“we value that perspective — take time to explore alternatives”) to maintain the behavior. If instead the response is “that’ll take too long, let’s just try the AI solution first,” that person will try 2–3 more times, then shut up — and you lose wisdom.
• Advice especially for higher-rank/leaders: align what you want with what you reinforce, lead by example, and be consistent/consequent — or the behavior slowly fades.
• Caveat: not disagreeing for the sake of it, and AI is often valuable — it’s about being able to have a meaningful disagreement and still agree to disagree.

5.2 Holding opposing views (cognitive dissonance)

• Fanni: humans are wired for black-and-white thinking (friend/foe survival), so it’s intellectually hard to hold two opposing arguments — and to see a disliked person as sensible and smart who also does good work.
• Evelyn: cognitive dissonance isn’t fun, but “this is exactly what we need in people” — the ability to make the uncomfortableness explicit and balance it, or polarization goes sky-high (“us vs. them”). We need people who can read these things and balance the technical and social aspects.

6. AI Bias and AI as an Authority

• AI didn’t make bias go away — it makes it more present, now with a confident writing style and an authoritative status in companies.
• The loop: we feed models biased data (because we’re biased) and expect unbiased output — instead it’s a reinforcing bias/AI loop (artifacts → model → suggestion → our context → our decisions → back in).
• AI is not rational either — it only learns from what we put in.
• Recent research: even when humans use the same sources as AI, we still treat an AI suggestion as an independent additional source — “fascinating,” bordering on dangerous.
• Anthropomorphizing: a session asked “who thinks Claude is a man?” — Evelyn raised her hand (“of course it’s a man, not a woman”). The point: in old socio-technical models the technical part was passive; now AI is a new, active actor creating novel output, which makes “how do we relate to it” far more complicated.

7. Engineering the Conditions (Social + Technical)

• “This is not the first disruptive wave, and the questions remain the same” — she doesn’t want to be the grumpy ”Statler and Waldorf” Muppet, but until we learn to handle the repeating questions (how does this fit our socio-technical system? how do we relate/optimize — socially and technically?), we’ll rehash this with the next wave.
• We should engineer the conditions — both social and technical — that produce the behavior we want. We’re busy engineering the technical conditions (token-maxing is a consequence), but not the social dynamics — even though the morning keynote’s examples were “very easy to explain if you take the social aspects into account.”
• We need more people who can optimize for both — tough, but fun.
• Never fully trust yourself either — fully trusting human intuition isn’t the way; the aim is balance.

8. Advice for Leaders

• There’s no one-size-fits-all framework. Example: companies say “we value deliberate thinking over AI-driven fast decision-making,” but deliberate thinking means something different in every team — you can’t blueprint it; it takes time, dedication, effort (and social/social-focused people) to figure out what it means in your context, then optimize for it. The blockers: time, money, motivation, willingness.
• You can’t change people, but you can change the environment — as a leader you engineer the conditions/environment.
• Have the conversation: what do we value? Then do critical self-reflection: is that what we’re reinforcing? If you say you value deliberate thinking but reinforce fast AI-driven decisions, it’s the leader’s responsibility to align them (“the gap between what you want and how you act”).
• Example from her earlier talk: asked “what behavior is reinforced in your company for decision-making?” the biggest answer was speed — then ask “is that what we think is most important?”
• The leader’s own behavior is an anchor — being higher in rank, “people look up to you literally,” so your behavior signals “how we do things around here.” Leaders must also notice what is being reinforced in them.

8.1 Power dynamics with AI

• Pre-AI, authority came from formal role, long tenure/experience, or being liked/trusted. Now AI is also seen as an authority — so what happens when a senior engineer and the AI disagree? “That’s not a technology question — it’s purely a question of power dynamics (politics and intuition).”

9. Closing

• Fanni’s summary: get curious and go into your fears; be honest, at least with yourself. If that’s hard, there are specialists who can facilitate it.
• And keep talking to people — look around and talk to others rather than only talking to the AI.

People & References Cited

• Evelyn van Kelle — guest; social scientist & change-management consultant; fan of heuristics.
• Fanni / Karolina Tóth — ABK Podcast / Leadership Anonymous host.
• Marian (Hartman) — quoted: “culture is a side effect of behaviors.”
• Concepts: human judgment (contextual/socially embedded), heuristics vs. frameworks (SAFe), heuristic journal, System 1, cognitive bias / debiasing, constructive/meaningful disagreement, cognitive dissonance, “you get what you reinforce,” AI bias / reinforcing bias loop, AI as authority / anthropomorphization (“Claude is a man”), socio-technical systems (active vs. passive tech), engineering social + technical conditions, token-maxing, power dynamics, culture as side-effect of behavior.
• Cultural references: Statler & Waldorf (the Muppets), Debrecen (Hungary).

Video: https://www.youtube.com/watch?v=KCq386YnaT8 — Transcript via yt-transcript.sh; outline generated from the transcript.