Four Live Q&As at Global Agile Summit 2026, Day Two

Day 2 had the same shape as Day 1 — recorded talks across four parallel tracks, plus a live Q&A for each presentation. The Q&As needed separate registration before each one started, and I had missed every one of Day 1’s because I did not know that step existed. On Day 2 I caught four, sequentially across an afternoon. Per Beining on Claude Code, Anna Lavrova on why systems resist change, Anita Kalmane-Boot on neurodiversity in software teams, Tara Scott on breaking the permission loop. The live Q&A surfaces what the speaker actually thinks about questions they did not prepare for — a different kind of substance than the recorded talk. This piece is about those four sessions; the recorded talks from Day 2 are a separate write-up.

I went in expecting four separate conversations on four different topics. What I came out with was four versions of the same observation. Anna stated it most directly when someone asked her about Hirschman’s loyalty / voice / exit / neglect framework: those four are not systems, those are the things rewarded by the system. The system itself is whatever reward structure produces those responses. Once she said it that way, the other three Q&As snapped into the same pattern. Per’s tokens, Anita’s school-system imprint, Tara’s permission-asking habit.

Per Beining — From Confused to Confident: How to Use Claude Code

Beginner-side conversation. Per started with Claude Code in January, three months ago; the questions came from people earlier on that arc. The advice was the kind of thing you can only give after building up actual hours: just go do, the population of people working with this stuff is doubling roughly every month so we are all early, the part you have to keep refreshing is the beginner’s mindset itself.

The token-consumption thread is what stuck with me. One participant said his tokens evaporate on the Pro tier and that the prepaid feel reminded him of dial-up internet from 25 years ago. Per’s response was an inversion: when your tokens are disappearing too fast, that is a trigger that you are ready for the next level. The next level is figuring out how to break the work down so the context does not balloon. He compared it to memory discipline on a 2KB ZX81 clone, his first computer. Same constraint, different decade. The constraint is doing what constraints are supposed to do.

I run into the same pattern with my own multi-agent pipeline for the big cats news project. The 2025 version of that pipeline runs up to 32 iterations of translation-quality loop across multiple models. Without aggressive context discipline — splitting the work into stages, each with its own narrow context — the cost would be untenable. Per’s framing of “tokens evaporating means you are ready to learn context discipline” matches what I actually had to do; I just had not named it that way.

The other line worth keeping: Per’s preference for the term “simulated intelligence” over “artificial intelligence” — picked up from a friend in conversation. AI implies the thing has the property; simulated intelligence keeps the framing honest. The naming will not change in public usage, but the framing matters internally.

What Per kept returning to was a shape similar to what Wallack and Greg Rog said on Day 1. Most of the failure modes are human and process problems wearing technical-problem masks. The bottleneck is judgement, not compute. Per’s version was softer and more practical: this is a new alien that has landed, you only learn what it can do by doing things with it, and the rest is rehearsal of patterns we already know from working with people.

Anna Lavrova — Why Systems Resist Change

Anna’s Day 1 talk walked through Hirschman’s loyalty / voice / exit / neglect framework on people in dysfunctional organisations. The Day 2 Q&A went somewhere else. I joined ten minutes late and walked into an exchange about whether all four responses are themselves systems. Anna’s answer reset the framing: those are not systems, those are the things rewarded by the system. The system itself is whatever reward structure produces those responses.

Then she told the 29-Batmans story. Engineering manager job, briefed by a CTO to “build a team out of these 29 engineers.” She walked in and found 29 senior, opinionated, deeply competent contributors who had never seen each other as teammates. The company’s pattern over ten years had been to pay them off whenever they raised an objection. Voice was rewarded with money, and the change being asked for was never executed. She compared the team to Avengers — except every member was Batman: a lone operator with gadgets who works alone and bullies anyone you put on a team with him. She left after four months because there was no team to build, only a golden cage full of expensive Batmans who could not collaborate and were too well-paid to leave.

The structural read is uncomfortable. Rewarding voice without acting on it is the same dysfunction as rewarding silence. Both shut down change. The 29-Batmans pattern is the failure mode of the “we listen to feedback” company that has not yet realised it stopped acting on any of it.

The other useful piece of the Q&A was on metrics, specifically velocity. Anna’s point: velocity is an illustration of speed, not a metric. The moment you turn it into a metric and reward against it, people optimise for the number. Three-point stories become five-point stories within a sprint. The team is not learning to work better; it is learning to demonstrate growth. Pete Oliver-Krueger added the counter-move: pair the metric with an opposing one, velocity plus predictability or velocity plus efficiency-of-flow, so neither can be gamed without breaking the other.

For me, the anti-pattern maps to my support team’s session-recording reviews of newly-registered users. The number we could chase is “issues identified per recording.” If we rewarded that metric in isolation, the team would learn to identify trivially small issues to keep the number up. We do not — we look at trends, weekly themes, what new users actually got blocked on — and the reason was something I had not articulated until Anna said it. The illustration is fine. The metric is the failure mode.

She closed with “make implicit explicit, make invisible visible.” Her example was a definition-of-done conversation where one team member kept naming things and the rest kept saying “yes but isn’t that common sense.” The act of writing it down was the actual change — not the content of the agreement, but the surfacing of agreement that had been silent. I have watched the same pattern in my own team. Surfacing silent agreement is its own piece of work, and it is easy to underrate because the content of the agreement does not change.

Anita Kalmane-Boot — Creating Space for Neurodiversity

The Q&A ran longer than the talk. The audience was self-selected — people who lived the topic in some form, themselves, a partner, or a child — and the conversation went somewhere I had not expected.

Two threads stuck. First, the “jerks are not autistic” piece. Anita’s friend started a website with that name to push back against bad behaviour being excused by neurodivergence. Neurodivergence does not erase the responsibility to behave well; treating it as if it does damages the case for accommodation by associating the case with bad actors. Jennifer, who is the parent of a neurodivergent child, added the parent angle: part of why the website exists is because lazy parenting under cover of “they are autistic” produces adults who never learn what is acceptable. The accommodations matter; the standards still apply. That is in voice with what I observe in support work.

Second, McKenna’s hero-syndrome question, where she described a team dynamic that looked like dominant-individual dysfunction and asked whether there was a neurodivergence connection. Anita’s answer was that the connection she sees most often is not about neurotype — it is about weak leadership. People who get away with hero-syndrome behaviour over a long period are usually being permitted to by managers who either do not recognise the cost or cannot intervene. The behaviour is not the cause; the absence of structure that would correct it is the cause. It is Anna’s reward-system point in different vocabulary.

I asked Anita the question I most wanted answered. How do you make the case for a neurodiversity initiative to a leadership group that has not asked for it and does not currently see why it matters? Asking it live is not natural for me; the question had been incubating since her talk and I had to force it into the room. Her answer pulled in three pieces.

First, know the person. Senior leaders who have not asked for input usually think they already know the answer. The framing that lands is whichever one fits their motivation, not yours — data if they are data-driven, story if they are narrative-driven, less-work-for-them if they are overloaded. Mismatch the framing and the conversation closes before it opens.

Second, you may not be the right messenger. Anita described the failure mode: someone too far from the leadership they are trying to reach — wrong relationship, wrong level — can have an unwinnable case from their seat regardless of how well it is built. Better to find someone the leadership actually listens to and brief them.

Third, decide whether the fight is worth it. Anita’s own example: she once proposed a neurodiversity-awareness session for her tech department and got back “why don’t we have a talk about how neurodivergents can work better with neurotypicals?” She named the response as missing the point, declined to spend energy turning it around, moved on. Some conversations cost more than they pay.

Pete added the quieter version. Leadership keeps moving — managers leave, get promoted, change reporting lines; people quit their managers or follow them when the manager moves. Whatever buy-in you build is bound to whoever sits in the chair, and the chair does not stay still. The implication is uncomfortable but useful: a case you painstakingly built into one person’s head is rarely transferable to the next person who arrives.

What I take from this is the explicit choice frame. The question is not always “how do I push harder for this.” Sometimes the answer is “this is not the conversation worth your time today, find a different one.” Both are valid. Treating the second as failure costs more than naming it as choice.

Anita closed on the pride-day pattern. The company that runs one neurodiversity event in a designated week and then carries on with unchanged hiring processes, unchanged interview formats, unchanged meeting structures. The performance is not the practice. The practice is what people do on Wednesday at 11am when no event is scheduled. I think this generalises past neurodiversity. Anything that is real about an organisation shows up in the ordinary working week, not in the calendar event.

The piece worth borrowing from Anita’s own practice is concrete. She walks new interns through what flexible working hours actually mean, what to do when stuck, when to ask, when to take breaks, what the company events are and that they are welcome at them. The implicit-to-explicit move that Anna talked about with definitions of done shows up here in another form. The rules everyone “knows” are not actually known by people who are new, neurodivergent, from a different educational background, or any combination.

Tara Scott — Breaking the Permission Loop

Tara’s Q&A came right after Anita’s. Same Pete and Alina hosting; mostly different audience. The opening question was about asking permission — when, when not. Her reframing was that there are three things people are usually conflating. Permission (“may I do this”). Consent (“I want to do this, here is the context, do you agree”). Direction (autonomy with enough context from a leader to act on). One of her workshop participants had recently named autonomy as his biggest challenge, not because he wanted to be told what to do, but because no one was there to help him, and the absence of direction had translated into permission-paralysis at the team level. Autonomy without direction is alienating. That is the part most “give people autonomy” advice misses.

The strongest piece of the Q&A was on micromanagement-as-fear. She told a story about an old senior director who called her at 11:30pm the night before quarter-end and asked her to skew the numbers in the story they were telling. They were close enough that she could ask why he was that scared. He retires in six years. His biggest stated fear is having a heart attack the day he retires. She left that job not long after. The point she was making was not about him. It was about the structural pattern. Micromanagement is the visible behaviour; underneath it, almost always, is a leader who is afraid of failing in a specific way and who has not been given a structure or relationship that lets them name the fear. The downstream cost is everyone below them learning to ask permission.

The third piece I want to remember is her response to a question Pete relayed from chat — what to do if you have 95% of the information but the personality type that focuses on the 5% you do not have. Tara typed herself as ENFP, the 16personalities campaigner, and named her partner as INTJ — a building architect who thinks in layouts, plans, and strategy. Different stack, different processing speed. Her partner has taken Tara’s classes and warned her up front: I am not going to ask questions, I need time to digest, I need to focus on what was said before I can respond. Tara took the warning at face value and adapted around it. Her advice in the Q&A was to sharpen the sword you actually have rather than try to perform the sword you do not. People in a workshop who stay quiet for forty minutes and then say one precise thing are doing the sword-sharpening. The contribution is not weaker for being delayed, but the room is often built to penalise the delay. That part is not the analyst’s problem to fix.

What ties the four together

Every one of these conversations spent its time on the same mechanism in different vocabulary. Per: tokens are a constraint that teaches you context discipline if the price signal is honest. Anna: organisations get the responses they pay for, and “we listen” without action is the same shape as “do not speak up.” Anita: weak leadership and pride-day theatre are the same failure mode, performance without a practice underneath. Tara: micromanagement is fear made structural, and permission-asking is what people learn when the fear is never named.

Anna’s reframe — those are not systems, those are the things rewarded by the system — turned out to be the day’s line. It works elsewhere. Run it on metric design, AI-tool adoption, parenting, permission norms. It keeps producing usable diagnostic questions. What does this system actually pay for? What behaviour has it trained without anyone noticing? What signal got sent that got optimised against?

Two things from this set will travel with me. The 29-Batmans story as a worked example of voice rewarded with money instead of action — useful next time I have to explain why “we have a feedback channel” does not by itself mean a system listens. And the implicit-to-explicit move as something to do more deliberately on my support team, especially around onboarding norms and the things everyone “knows.”

My Day 1 piece covered thirteen recorded talks back to back. This one is four live Q&As across an afternoon, with the questions doing most of the work. Two different cuts of the same conference. The pattern underneath was the same.