AI clarity session or build-with-me session?
How to decide between exploration and concrete building. Signals, outputs, agendas, what goes wrong when you pick the wrong format, and when to combine both.
Key takeaways
- Choose the session type based on where the uncertainty sits.
- Use clarity sessions when the opportunity is vague and build sessions when the artefacts are already visible.
- Both formats should end with something practical enough to test.
It is about the shape of the uncertainty, not seniority
People sometimes assume clarity sessions are for beginners. They are not. The uncertainty is just a different shape. Fog about the opportunity space is not the same as fog about implementation detail.
An executive drowning in contradictory vendor decks still needs to narrow before commissioning anything. A practiced operator wrestling a gnarly document pipeline probably needs hands-on prototyping, not another landscape slide.
Pick the format that matches where the fog actually sits.
Clarity sessions: what they solve
They take a dozen half-baked frustrations and converge them into a sharper map. Where AI realistically helps. Where to leave things manual for now.
Outputs look like portfolio thinking. A list of candidate workflows, rough impact-vs-ease scoring, prerequisites (data cleanliness, ownership), likely failure modes, and an honest sequence across quarters rather than afternoons.
They still have to touch the ground. Bring role context, sanitised artefacts, and calendar realities. Philosophy alone does not sharpen anything.
Build-with-me sessions: what they solve
When you can show the artefacts, the CSV export, the brittle macro, the draft email pattern, the time gets spent co-editing. Live iteration surfaces hidden constraints dashboards never recorded. Tone, edge cases that lurk in row seventeen, brittle integration assumptions.
Concrete things come out. A scaffolded prompt suite. A repeatable checklist transformation. A minimum viable automation. All measured against rework minutes saved this week, not against future productivity gains nobody will reconcile.
The momentum depends on who is in the room. The person who owns the follow-up tweaks has to be there, or has to delegate authority live. Otherwise the artefact fossilises.
Agenda sketches (so expectations stay honest)
Clarity (around 120 minutes): context intake, inventory candidate loops, score them on impact and confidence, converge on the top few experiments with named owners, leave with a roadmap and checkpoints.
Build-focused (often longer, contiguous): agree on success criteria upfront, dissect sample input and output pairs, run iterative prompt and design cycles with recorded examples, define the review loop and ownership, decide what to instrument next (evaluation metrics or scripted checks).
Hybrid sequences are deliberate. Coarse clarity sprint on Friday, focused build sprint the following week once the homework, cleaner samples, has landed.
Mismatched formats hurt outcomes
Asking for a clarity session when the artefacts are already there leads to abstract reinvention. You feel stalled while someone "strategizes" a mess you already know intimately.
Rushing into a build before the strategic alignment is done leaves brittle artefacts nobody maintains. Beautiful prompt libraries nobody trusts, because the organisational permissioning never matured.
If you spot the mismatch, call it out early. Saying out loud “we should swap modes” saves everyone calendar debt and carries no stigma.
Bridging out after either session
Clarity exits should land on the next tactile deliverable, not on lofty OKRs. Name the artefact skeleton, the stakeholder who signs off, and the rehearsal window. Even if the build sits in the next sprint.
Build exits should freeze the version openly. Versioned prompts. Pinned model snapshot references if reproducibility matters legally or ethically downstream.
Budget emotional runway too. Behavioural adoption usually lags technical readiness. Pair the scaffolding with short enablement notes coworkers can copy without joining the workshop.
Revisit quarterly. The AI surface shifts. Organisational memory decays faster than codebase comments. A lightweight retro keeps the map honest.