Question 1

What is the difference between coaching, consulting, mentoring and teaching, and where can AI genuinely help?

Accepted Answer

Consulting gives an answer, mentoring shares experience, teaching transfers knowledge, and coaching helps someone reach their own answer. Most of my work moves between all four depending on what the person actually needs. AI is genuinely useful for the teaching layer — generating practice material, answering framework questions, building learning paths — and increasingly for consulting-style analysis. It is close to useless for coaching, because coaching depends on a relationship in which someone is willing to be uncertain in front of you. That willingness is not something a tool can be granted.

Question 2

The board has mandated an AI-first transformation in twelve months. What do you do in your first ninety days?

Accepted Answer

I would spend the first month finding out where value actually gets stuck, because an AI-first mandate is a solution looking for a problem until you know that. That means going to where the work happens: how long an idea takes to reach a customer, where it waits, what the rework rate is, and which of those delays a language model could plausibly touch. Usually the answer is that the constraint is decision latency or handoffs, which no tool fixes. Then I would pick two or three areas where the case is real, run them properly with a baseline, and give the board evidence at ninety days rather than a maturity assessment. The risk I am managing is that the mandate turns into tool deployment counted as progress.

Question 3

Individual developer output has risen sharply since the AI rollout, but time to market has not improved at all. What is happening?

Accepted Answer

You optimised a non-constraint. Writing code was never the bottleneck in most organisations — the delays live in review, testing environments, approvals, release windows, and the wait between teams. Speeding up the one step that was already fast just fills the queues in front of everything else faster. So I would map where the time actually goes end to end, show leadership that picture, and direct the next investment at the real constraint, which is usually verification capacity or an approval process nobody owns. It also reframes the conversation productively: the tools are working exactly as sold, and the system is what is limiting the return.

Question 4

Does the availability of AI agents change how you would design teams?

Accepted Answer

It changes team sizing more than team structure. The principles that made small, cross-functional, long-lived teams work are about communication cost and ownership, and agents do not reduce either. What does change is the internal balance: a team that can generate far more change than it can review needs more reviewing and integrating capability than it used to, so the shape shifts toward senior verification skill rather than toward more generators. The other change is that platform and enablement teams get more valuable, because guardrails, evaluation infrastructure and safe defaults are now what determines whether generated work is usable.

Question 5

The CIO wants to cut delivery headcount by twenty percent, citing AI productivity gains. You are asked to support the plan. What do you do?

Accepted Answer

I would ask what evidence the twenty percent is based on, because in my experience it comes from vendor material rather than from their own delivery data. Then I would put our own numbers next to it: throughput and lead time before and after, and where the remaining wait time sits. Two specific risks belong in that conversation. The capability the cut usually removes first is reviewing and mentoring, which is exactly the capacity a generating organisation is short of. And the people who leave voluntarily during a cut are the ones with options, who are your strongest. If the organisation needs to reduce cost, I would say so honestly and help do it well, but I would not let AI productivity be the justification unless the data supports it.

Question 6

How would you measure the return on an AI-augmented transformation?

Accepted Answer

I would insist on a baseline before anything changes, because without it every later number is an argument rather than evidence, and this is the step organisations skip. Then I would measure at the level the business cares about — time from idea to customer value, quality in production, and cost to serve — with tool usage kept firmly as an input, not a result. Where possible I would keep a comparison group of teams who adopt later, which is the only practical way to separate the effect of the tools from everything else changing at the same time. I would also insist on reporting what got worse, because a transformation report with no negative findings is not being read by anyone serious.

Question 7

Leadership wants an AI-generated agility maturity score for forty teams, updated monthly. How do you respond?

Accepted Answer

I would ask what decision the score informs, and usually there is not one — it exists so that leadership feels informed. The moment a score is published and compared, teams optimise for the score, and within two quarters you have expensive, well-decorated data telling you nothing about how the work actually flows. What I would offer instead is a small set of outcome measures teams cannot easily game because they come from the work itself, plus qualitative depth from actually visiting teams. If leadership genuinely needs a portfolio view, I would rather give them four honest numbers than forty confident ones.

Question 8

Two years into a transformation, the events run and the tools are deployed, but nothing about how decisions get made has changed. What now?

Accepted Answer

I would stop adding practices, because more practice on top of an unchanged decision system is what created the theatre. The honest diagnosis is usually that the organisation adopted the visible parts, which are cheap, and skipped the expensive part, which is moving decision authority closer to the work. So I would take one real constraint — a funding cycle, an approval gate, a hiring decision — and try to change that single thing with a sponsor who has the authority. One genuine change in how decisions are made teaches more than another year of ceremonies. If no executive is willing to change any decision right, I would say plainly that the engagement is not going to succeed and let them decide whether to continue.

Question 9

Would you use AI to prepare for or reflect on a coaching conversation?

Accepted Answer

For preparation, yes, and it is genuinely useful: rehearsing a difficult conversation, generating questions I would not have thought of, or pressure-testing my own framing before I walk in. What I would not do is put a real person's identifiable situation into a tool. What someone tells me in a coaching conversation is theirs, and putting it into a system I do not control breaks the confidentiality the relationship depends on, whatever the vendor's data policy says. So I abstract to the pattern rather than the person: I can ask about handling defensiveness in a senior leader without describing which leader.

Question 10

If AI absorbs the administrative half of Scrum Master and coach work, what happens to those roles in an organisation you are advising?

Accepted Answer

Roles that were mostly administration will not survive, and I would not pretend otherwise. What that exposes is a distribution that was always there: a portion of people in these roles were coordinating and reporting, and a portion were changing how the organisation works. The second group becomes more valuable, not less, because there is now more change to absorb and more risk to manage. So my advice to an organisation is to stop treating it as a headcount question and start being explicit about which work they are buying, and my advice to the people in those roles is to move deliberately toward the part that requires standing behind a judgement.

Question 11

The organisation's governance runs on stage gates and up-front estimates. AI work does not fit that shape. How do you change it?

Accepted Answer

I would not argue that gates are wrong, because governance exists for a legitimate reason and the people who own it are protecting something real. I would ask what each gate is actually protecting against — usually unrecoverable spend, compliance exposure, or a commitment made to a customer — and then propose a way to protect the same thing with evidence rather than a document. For work with uncertain feasibility, that means funding a decision point rather than a deliverable: a bounded amount of money to answer a specific question, with pre-agreed criteria for continuing or stopping. That is a language finance understands, and it survives after I leave, which a coaching intervention does not.

Question 12

An executive tells you Agile is dead and AI has made it obsolete. Respond.

Accepted Answer

I would concede the part that is true. A great deal of what was sold as Agile was ceremony and certification, and if that is what they mean, they are right and I would not defend it. What is not obsolete is the underlying problem: when you cannot predict what will work, you have to build in small pieces, put them in front of real users, and change course on evidence. AI makes that more necessary rather than less, because you now cannot even be sure a feature is technically feasible until you have tried it. So the empirical part matters more, and the ritual part matters less. If their experience of Agile was mostly ritual, that is a fair criticism of how it was implemented here.

Question 13

How do you roll out AI tooling across thirty teams without mandating it?

Accepted Answer

Mandates get you licence activation and nothing else. I would start with the teams who already want it, help them do it well, and make sure the outcome is visible and honest, including what did not work. Then I would remove the friction the next group will hit: approved tooling, a clear data boundary, working examples in their own codebase, and someone to ask. A community of practice where teams share what actually helped does more than a rollout programme. The one thing I would mandate is the boundary, because data protection is not a preference. Everything past that boundary is theirs to choose.

Question 14

You discover widespread unapproved AI tool use on personal accounts across the organisation. What do you do?

Accepted Answer

I would treat it as a signal before treating it as a violation. People went around the process because the sanctioned path was too slow or did not exist, and punishing that first drives it further underground where you have no visibility at all. So the immediate action is a fast, credible, approved alternative, communicated clearly, together with a plain statement of what must never leave the organisation. Then I would deal with genuine exposure that has already occurred, honestly and through the right channels. The lasting fix is shortening the approval path, because if it takes six months to sanction a tool, this will happen again with the next one.

Question 15

Teams can generate far more than they can verify. How would you build the missing capability across an organisation?

Accepted Answer

Not with a course, because reviewing skill is built by reviewing under guidance, not by attending. I would put the learning inside the work: pairing on reviews, senior engineers walking through what they look for and what they missed, and short internal sessions built from the organisation's own recent incidents, which are far more persuasive than generic material. Alongside that I would invest in the things that reduce how much verification is needed at all — better automated testing, stronger guardrails, smaller batches. The measure I would watch is defect escape rate and rework, not course completions.

Question 16

HR asks for your help using delivery and AI-usage data to identify low performers. What do you say?

Accepted Answer

I would decline to help build that, and I would explain why in terms of what it will cost them rather than in terms of ethics alone. Delivery data measures a system, not a person: an individual's numbers are dominated by what they were assigned, who they were unblocking, and how the work was split. Ranking on it reliably punishes the people doing the invisible work that holds teams together, and once people know they are ranked this way, the data stops describing reality within one quarter. If there is a genuine performance concern about an individual, that belongs with their manager, based on observed work and outcomes, and I would help them do that conversation well.

Question 17

A client asks you to guarantee a thirty percent productivity improvement from AI adoption as part of the engagement. How do you respond?

Accepted Answer

I would not sign that, and I would say why rather than just refusing. Nobody can guarantee an outcome that depends on decisions the client controls — their release process, their approval chain, their willingness to change how work is funded. What I can commit to is measurable and inside my control: a baseline established in the first month, specific constraints addressed, and a decision point at which they can stop. If they want an outcome-based commercial arrangement, I would rather tie it to something we can jointly influence and measure, with the client's own obligations written into it. A guarantee I cannot keep loses the client anyway, six months later and more expensively.

Question 18

Your organisation sells fixed-price delivery. AI is changing the cost base. How should the commercial model adapt?

Accepted Answer

The uncomfortable part is that billing for effort while effort falls is a shrinking business, and clients are working this out. The direction that survives is charging for outcomes and for risk transfer rather than for hours, which means the supplier keeps the productivity gain in exchange for carrying delivery risk. That only works with real measurement and disciplined scope, and it exposes suppliers whose margin came from headcount rather than capability. I would also expect the mix to shift toward the work that is now scarcer: verification, integration, and accountability for what was generated.

Question 19

A leader you coach presents polished AI-generated strategy documents with no real thinking behind them. How do you handle it?

Accepted Answer

I would not comment on how the document was produced, because that turns it into an accusation and ends the coaching relationship. I would engage with the content, in private, by asking the questions the document does not answer: what would have to be true for this to work, what did you decide against, who disagrees with this and why. That surfaces the gap without me naming it, and it lets them arrive at the problem themselves, which is the only version that changes anything. The underlying issue is usually that they are producing artefacts because artefacts are what gets rewarded here, and that is worth raising separately with whoever set that expectation.

Question 20

The organisation funds annually against fixed business cases. Much AI work cannot be specified that far ahead. How do you change the funding model?

Accepted Answer

I would work with finance rather than campaigning against annual budgeting, which is usually driven by obligations they cannot unilaterally drop. The practical move is to change what the money buys inside the annual envelope: fund a stable team against a problem for a period, rather than a specification, and review continuation on evidence at set points. That keeps the annual commitment finance needs while removing the requirement to specify the unspecifiable. I would start with one portfolio area, prove the reporting works, and let it spread on its own merits. Attempting to change enterprise funding by argument alone does not work.

Question 21

You are facilitating a two-hundred-person session. How do you use AI, and where does it mislead you?

Accepted Answer

It is genuinely valuable for the part that used to force a break in the session: clustering hundreds of written contributions into themes in minutes so the group can react to them while still in the room. Where it misleads is that it flattens toward the majority. The outlier contribution from the one person who understands a risk nobody else sees gets absorbed into a nearby theme and disappears, and that contribution is frequently the most valuable thing in the room. So I read the raw input myself for outliers before presenting the themes, and I show the clusters as a draft the group corrects rather than as the output of the session.

Question 22

If verification has become the organisational constraint, what changes outside the delivery teams?

Accepted Answer

Quite a lot, and most of it sits outside engineering. Hiring shifts toward senior judgement rather than volume, which changes what recruitment is asked to source. Career paths need a route that rewards reviewing and mentoring, because today those are unpaid work that slows your own delivery numbers. Quality and security functions move from end-of-line inspection to building the automated gates that reduce human review load. Procurement has to evaluate tools on what they do to verification cost, not just generation speed. And leadership has to stop reading output volume as progress, which is the hardest of the five.

Question 23

How do you build a community of practice around AI use that does not degenerate into tool worship?

Accepted Answer

By making the currency evidence rather than novelty. A demo session where people show impressive outputs produces enthusiasm and no learning. What works is a regular slot where people bring something they tried on real work, with what it cost, what it produced, and whether they kept doing it — including the things they abandoned, which are the most useful contributions and the hardest to get people to share. I would also keep a short written record of what the organisation has learned, because otherwise the same experiment is run five times by five teams. And I would deliberately invite the sceptics, since a community that only contains enthusiasts stops being able to evaluate anything.

Question 24

Middle managers are the main source of resistance. They are also the ones whose roles look most threatened. How do you work with them?

Accepted Answer

I would start by accepting that their resistance is rational. They are usually being asked to give up control of how work is done while remaining accountable for its outcome, and now they are also hearing that AI reduces the need for coordination, which is what many of them do. Telling them to trust the process is not an answer to that. What has worked for me is being concrete about what their role becomes — removing the impediments teams cannot reach, developing people, and owning outcomes rather than activity — and then making sure the organisation actually rewards that, because if it does not, they are right to resist. Where the honest answer is that fewer of these roles are needed, they deserve to hear it early rather than discover it.

Question 25

A team asks you to just tell them which AI tools to use. Do you?

Accepted Answer

Yes, mostly. Withholding an answer I have in order to preserve a coaching stance is self-indulgent when the team simply needs the information to get on with their work. I would give a direct recommendation with the reasoning behind it, and be clear where I am uncertain. What I would not do is decide for them where the decision has real consequences for how they work, or where the choice belongs to them and they are trying to hand it back. The distinction I use is whether the question is about information or about ownership.

Question 26

You joined after the AI rollout, so there is no baseline. How do you prove anything?

Accepted Answer

I would reconstruct what I can from history that already exists — work-tracking timestamps, deployment records, incident history — which usually gives a usable picture of lead time and quality for the period before the rollout even if nobody was measuring deliberately. Then I would be explicit about its limits rather than presenting it as clean evidence. From today forward I would establish a proper baseline for the next change, and where teams are adopting in waves I would use the later waves as a comparison. Above all I would resist manufacturing a retrospective improvement number, because those get quoted for years and eventually someone checks.

Question 27

A vendor has pitched an AI agile management platform directly to your sponsor, who is enthusiastic. How do you handle it?

Accepted Answer

I would not attack the product, because that becomes a contest with the sponsor rather than with the claim. I would ask what problem they expect it to solve and then look for evidence that the problem is a tooling problem at all — most of these platforms automate reporting, and reporting is rarely the constraint. I would propose a bounded trial with one or two teams, with success criteria agreed in advance, including what would make us walk away. That usually settles it on evidence in six weeks. I would also raise the questions vendors dislike: what happens to our data, what the exit path looks like, and what the cost is at full scale rather than at pilot scale.

Question 28

Teams are producing more, and leadership has quietly raised what it expects. People are showing signs of strain. What do you do?

Accepted Answer

I would treat it as a delivery risk rather than a wellbeing footnote, because that is the framing that gets it addressed. What has usually happened is that the expectation absorbed the entire gain, so the same pressure now sits on top of a higher output, and the additional load is cognitive: reviewing generated work is more tiring than writing your own and it is invisible in every measure leadership looks at. I would make that load visible, show the trend in defects and rework that follows sustained strain, and be direct with leadership that a pace nobody can hold is a plan to lose your senior people in nine months. Then I would push for some of the gain to be reinvested in the work itself rather than all of it in output.

Question 29

How do you know when your coaching engagement should end?

Accepted Answer

When the organisation is solving problems I used to be called into, and my absence from a week does not change anything. I try to make that explicit at the start by agreeing what should be true when I leave, so ending is a planned outcome rather than a budget decision. I also watch for the failure mode in myself: if I am still the person who runs the important sessions or breaks the deadlocks after a year, I have created dependency and I should be reducing my presence, not defending it. Staying longer than useful is an easy mistake to make when the client is happy and the invoicing is comfortable.

Question 30

Tell me about a transformation you worked on that failed.

Accepted Answer

Structure it as the situation, what you were brought in to change, what actually happened, what you now believe you got wrong, and what you do differently because of it. Pick a real failure — an engagement that ended early, a change that reverted, sponsorship you never secured — and locate your own contribution in it rather than the client's dysfunction. The strongest version names something you would recognise earlier now, such as a sponsor who agreed in the room and never changed a decision. Avoid the failure that turns out to have been a triumph of learning.

Agile Coach Interview Questions and Answers for the AI Era

What is the difference between coaching, consulting, mentoring and teaching, and where can AI genuinely help?

A strong answer

Answers that lose you the room

The board has mandated an AI-first transformation in twelve months. What do you do in your first ninety days?

A strong answer

Answers that lose you the room

Individual developer output has risen sharply since the AI rollout, but time to market has not improved at all. What is happening?

A strong answer

Answers that lose you the room

Does the availability of AI agents change how you would design teams?

A strong answer

Answers that lose you the room

The CIO wants to cut delivery headcount by twenty percent, citing AI productivity gains. You are asked to support the plan. What do you do?

A strong answer

Answers that lose you the room

How would you measure the return on an AI-augmented transformation?

A strong answer

Answers that lose you the room

Leadership wants an AI-generated agility maturity score for forty teams, updated monthly. How do you respond?

A strong answer

Answers that lose you the room

Two years into a transformation, the events run and the tools are deployed, but nothing about how decisions get made has changed. What now?

A strong answer

Answers that lose you the room

Would you use AI to prepare for or reflect on a coaching conversation?

A strong answer

Answers that lose you the room

If AI absorbs the administrative half of Scrum Master and coach work, what happens to those roles in an organisation you are advising?

A strong answer

Answers that lose you the room

The organisation's governance runs on stage gates and up-front estimates. AI work does not fit that shape. How do you change it?

A strong answer

Answers that lose you the room

An executive tells you Agile is dead and AI has made it obsolete. Respond.

A strong answer

Answers that lose you the room

How do you roll out AI tooling across thirty teams without mandating it?

A strong answer

Answers that lose you the room

You discover widespread unapproved AI tool use on personal accounts across the organisation. What do you do?

A strong answer

Answers that lose you the room

Teams can generate far more than they can verify. How would you build the missing capability across an organisation?

A strong answer

Answers that lose you the room

HR asks for your help using delivery and AI-usage data to identify low performers. What do you say?

A strong answer

Answers that lose you the room

A client asks you to guarantee a thirty percent productivity improvement from AI adoption as part of the engagement. How do you respond?

A strong answer

Answers that lose you the room

Your organisation sells fixed-price delivery. AI is changing the cost base. How should the commercial model adapt?

A strong answer

Answers that lose you the room

A leader you coach presents polished AI-generated strategy documents with no real thinking behind them. How do you handle it?

A strong answer

Answers that lose you the room

The organisation funds annually against fixed business cases. Much AI work cannot be specified that far ahead. How do you change the funding model?

A strong answer

Answers that lose you the room

You are facilitating a two-hundred-person session. How do you use AI, and where does it mislead you?

A strong answer

Answers that lose you the room

If verification has become the organisational constraint, what changes outside the delivery teams?

A strong answer

Answers that lose you the room

How do you build a community of practice around AI use that does not degenerate into tool worship?

A strong answer

Answers that lose you the room

Middle managers are the main source of resistance. They are also the ones whose roles look most threatened. How do you work with them?

A strong answer

Answers that lose you the room

A team asks you to just tell them which AI tools to use. Do you?

A strong answer

Answers that lose you the room

You joined after the AI rollout, so there is no baseline. How do you prove anything?

A strong answer

Answers that lose you the room

A vendor has pitched an AI agile management platform directly to your sponsor, who is enthusiastic. How do you handle it?