Haven't taken the check yet? Do that first — 14 questions, about four minutes. It scores you on the three areas below and tells you which one to fix first. Already have your result? Jump to your weakest area: Critical thinking · Sustainable pace · Psychological safety.

01The model

Three things to protect. Strong thinking that nobody dares voice is wasted; a safe team with an unmanaged pace just shares the strain out loud. Scope note: this is team practices and process — anything that needs individual care belongs with a qualified professional.

01Critical thinking — AI as a draft to challenge, not a result to accept. Someone still reasons from first principles.
02Sustainable pace — the time AI frees up is spent by choice — and there’s protected time to think.
03Psychological safety — "this AI output is wrong" and "we’re going too fast" can be said without cost.

The check measures how the team works — not how anyone is doing. The engagement, in a sentence: find the weak one and reinforce it first.

02Critical thinking

The test: can people explain and defend AI-assisted work as their own reasoning? If the honest answer is "the model said so", judgement has been outsourced.

What good looks like
  • AI output is a draft to challenge, not an answer to accept.
  • Someone still works the hard problem through from first principles — it's protected time.
  • People are expected to explain and defend AI-assisted work as their own.
  • The team has named the judgement calls it won't delegate to a model.
  • Reasoning gets deliberate practice — not just prompting technique.
Common failure
  • Outsourced judgement — nobody can defend the reasoning behind the work.
  • Prompt, paste, ship — no step where a human reasons it through.
  • Only seniors check — juniors watch AI output go out uncorrected.
Your first move
If you're just starting

Make "explain it as your own" the standard for anything that matters — if you can't defend the reasoning, it isn't done. Add a norm that AI output is a draft to challenge, and set aside time where the team works a problem before it prompts.

If it's partly there

The team challenges AI output for big things — extend that habit down, and name the judgement calls it should own outright. A short standing conversation about reasoning (not prompting technique) keeps the muscle in use.

If it's already strong

Keep the "defend it as your own" bar high as deadlines press, keep naming what the team should not delegate, and watch for first-principles work quietly disappearing from the calendar.

Go deeper

  • Add one line to review: “what’s the reasoning here, in your words?” — asked of AI-assisted and hand-written work alike.
  • Write a short list of calls the team won’t delegate to a model — pricing, hiring, anything irreversible — and keep it visible.
  • Run a 30-minute “work it before you prompt it” session on a real problem each month.

03Sustainable pace

Pace is whether the time AI frees up is spent by choice. The test: has the team actually decided what that time is for — and protected any of it?

What good looks like
  • The team has decided what the time AI frees up is for — and protected some of it.
  • There's real, designed-in time for focused, uninterrupted work.
  • Plans account for the true cost of context-switching and AI-wrangling.
  • Workload and pace are reviewed as a team, not left to each person.
  • The team resists false urgency — AI-speed doesn't make everything urgent.
Common failure
  • Throughput creep — “AI makes us faster” became “do more”, silently.
  • Silent overload — everyone managing their own strain, alone.
  • Everything urgent — AI-speed expectations on every request.
Your first move
If you're just starting

Have the conversation the team hasn't had: what is the time AI frees up actually for? Then protect some of it. Design deep-work time into the week, and plan with the real cost of context-switching in view — not the demo-speed version.

If it's partly there

Pace is discussed but not yet designed. Make workload a standing item in retros, protect focus time properly rather than in theory, and give the team language to push back on false urgency — most "urgent" AI-era asks aren't.

If it's already strong

Keep reviewing workload as a team, keep deep-work time defended when demand rises, and keep separating urgent from important — that discipline erodes fastest under pressure.

Go deeper

  • Put one protected deep-work block per person per week on the calendar and defend it like a meeting.
  • Add “how’s the load?” as a standing retro question, answered by the team, not privately.
  • Agree a simple test for “urgent” — what breaks if this waits a day? — and apply it to the next five “urgent” asks.

04Psychological safety

Safety here is narrow and specific: can anyone say "I don't know", "this AI output is wrong", or "we're going too fast" without it costing them?

What good looks like
  • “I don't know” and “I didn't check that with AI” carry no penalty.
  • Flagging that an AI output is wrong or biased is welcomed and acted on.
  • “We're moving too fast” or “this shortcut worries me” gets heard — from anyone.
  • When AI-assisted work goes wrong, the response is a blameless look at the process.
Common failure
  • Blame the prompter — when something breaks, find who typed it.
  • Only seniors speak — juniors watch the AI output go out uncorrected.
  • Silence on pace — “too fast” never gets said, only felt.
Your first move
If you're just starting

The fastest fix is how the team responds to problems: make it blameless and about the process, visibly, the next time something goes wrong. Then make "I don't know" and "this AI output looks wrong" normal things to say — model it from wherever the authority sits.

If it's partly there

People speak up sometimes — make it reliable. Act visibly on the next "this is wrong" or "we're going too fast", so raising it is seen to be worth it, and check that it isn't only the senior voices who feel able to.

If it's already strong

Keep acting on the hard flags, keep the blameless response under deadline pressure, and keep checking that the newest and most junior people feel as able to speak as anyone else.

Go deeper

  • The next time AI-assisted work goes wrong, run the review on the process out loud — no names — and share what changed.
  • Ask the most junior person in the next review for their read first, before anyone senior anchors it.
  • Add “anything we’re rushing?” to planning, and act on the first honest answer so it’s seen to be safe.
🛡️

Guardian AI doubles as a critical-thinking checkpoint — it asks whether a human actually thought about a prompt before it's sent. Free public beta.

05Do this in the next two weeks

  • Take (or retake) the team check and note which of thinking, pace or safety scored lowest.
  • For thinking: add "what’s the reasoning, in your words?" to review, and list the calls you won’t delegate to a model.
  • For pace: put one protected deep-work block per person on the calendar, and add "how’s the load?" to the retro.
  • For safety: run the next incident review blameless and on the process, out loud, and share what changed.
  • Agree a one-line test for "urgent" and apply it to the next five urgent asks.
  • Pick the single weakest area, put a named owner and a date on its first move, leave the other two as "next".

Still stuck?

If the weak area won't move — the reasoning habit keeps thinning, pace creeps back up, people still won't say "wrong" out loud — that's the point to bring Jon in on that one problem. A facilitated session with the team on how thinking, pace and candour actually play out, then the move: a working agreement, a workshop, or a change to how the team plans. Team practice only — individual support is a separate, professional matter.