How the leader track works
Six sessions, and they are about decisions, not demos. You will not be asked to open a tool. You will be asked to decide things that only you can decide, and to decide them with numbers rather than vibes. Tonight: what genuinely changed in the work, and where the money now leaks. a2 gives you a critique standard you can install in a review, so that half of taste stops being a debate. a3 is the honest tool and workflow map - what each vendor actually outputs, and which seat or plan gates it. a4 is IP and provenance: what you can register, what a vendor will and will not indemnify, and what C2PA does and does not prove. a5 is craft and careers - where juniors used to learn, and what replaces the rung AI just removed. a6 is rollout: the metrics, the guardrails and a twelve-month plan you could take to a board.
Running product across both tracks is Cadence, an AI meeting-notes app, and the running artifact is its shared-summary feature. The designer track (b1-b10) takes that feature from brief to a shipped, spec'd screen set, and it does so against a live critique lab that scores a real screen on real measurements. You will hear its numbers quoted in a2 - the climb from 17 to 100 - because a stated standard is a leadership tool, not a designer's toy. You do not need to run the lab to lead the conversation.
Where the cost actually collapsed 8 min live
Start with the parts that are real, because they are large. Exploration volume collapsed: twenty directions costs roughly what one used to. First drafts collapsed: the blank artboard, the placeholder layout, the "something to react to" is close to free. Asset variants collapsed: background out, object removed, image extended, resolution up. Resizing and adaptation collapsed: one layout into nine placements is a queue, not a week. Boilerplate copy collapsed: the shorter version, the softer tone, the eleventh translation. If your team is still spending human hours on that list, that is a process problem now, not a staffing one.
Now the parts that did not move, and notice they are all the same kind of work. Deciding which option is right did not move; forty options is a harder decision than four, not an easier one. Brand-specific judgement did not move, because a generator returns what is likely and your brand is by definition not the likely thing. States and edge cases did not move - empty, loading, error, one item, four hundred items, the summary that failed to generate. And accountability did not move at all. Nobody has ever escalated to a model.
LiveThe bottleneck moved from making to deciding4 min▶
This is the single sentence to take into your next planning meeting. When making was expensive, the org optimised for making: headcount, tool licences, production pipelines, "can we get this out by Thursday". Everything downstream of production was under-resourced because production ate the calendar.
Now production is cheap and the queue has moved one step to the right. Watch for the symptoms in your own team:
- Option pile-ups. Reviews arrive with fifteen directions and no recommendation. The designer did more work and made your job harder.
- Decision latency. The slowest thing in the cycle is now a person with authority finding an hour.
- States debt. Beautiful happy paths, no empty state, no error, no long-content case. Generators do not volunteer the unglamorous screens.
- Reopened decisions. Because regenerating is free, "let's just try another version" replaces "we decided".
The leadership response is not more generation and it is not less. It is a decision standard, a review ritual, and a rule that options arrive with a recommendation attached. a2 builds the standard; a6 installs the ritual.
Self-studyThe four things you cannot delegate to a generator3 min read▶
Keep this list somewhere you can point at it, because it is what your team is actually paid for now.
| The work | Why AI cannot take it |
|---|---|
| Choosing the right option | A model can rank against a generic notion of quality. It cannot know that this quarter you are defending trust, not chasing delight. |
| Brand-specific judgement | Generators return the likely. A brand's whole job is to be recognisably not the average. The gap is structural, not a prompt problem. |
| States and edge cases | The failure screens are where product credibility lives, and they are exactly the screens that were never in the training data in any volume. |
| Accountability | When the summary leaks the wrong meeting, a named human answers for it. That is not a technical limitation you will engineer away. |
Notice that three of the four are judgement and one is governance. That is the shape of a modern design lead's job description.
You bought seats. You are spending credits 10 min live
Here is the trap. You approved seats, because seats are what design tools have always cost: a predictable number per person per year, negotiated once. But the AI features are not priced like seats. They are metered, in credits, per person, per month, with no rollover and no sharing across the team. Then you stood up in an all-hands and told everyone to explore more. Exploration is generation. Generation is the metered thing. You just increased consumption of the one line item that was not in the plan.
The numbers below were fetched from official vendor documentation on 8 August 2026, and every one of them is the kind of number that changes monthly. Treat them as a worked example of the shape of the problem, then re-check your own tenant and contract before you commit a budget.
- Figma AI credits are allocated per Full seat per month: Starter 500 with a 150 per day cap, Professional 3,000, Organization 3,500, Enterprise 4,250. No rollover, not shareable. Some actions are free (AI search, rename layers), a background removal is a few credits, and Figma Make runs 30 to 100 or more credits per task - one afternoon of prototyping can be a meaningful slice of a month. Pay-as-you-go exists at $0.03 per credit with a spend cap.
- Adobe generative credits renew monthly. Most standard features cost 1 credit per generation; premium features cost far more. Reported plan tiers include Creative Cloud Pro at 4,000 per month while new single-app plans get 25 per month - a range wide enough that "we have Creative Cloud" tells you almost nothing about capacity.
- Canva moved to a shared credit pool across three quality tiers - Standard, Premium and Ultra. Free gets 200 Standard or 20 Premium, and no Ultra at all. Credits are per member and never pooled across the team.
- Midjourney Basic at $10 buys 3.3 fast GPU hours and no Relax mode, so when the fast time is gone, it is gone until the next cycle. Fast time does not roll over.
LiveHow to price a month of your team's exploration4 min▶
You do not need a model. You need four numbers and one honest hour with the people doing the work.
Count the generating tasks. Per designer, per week: image generations, background removals, prototype builds, resizes, copy rewrites. Ask them to log a real week, not to estimate.
Attach a cost band to each. Cheap actions are a credit or a few. Prototype builds are the expensive tail - Figma Make alone runs 30 to over 100 credits per task.
Multiply to a month, then compare to the seat's allowance - not to a plan name. A Professional Full seat has 3,000 Figma AI credits a month, an Enterprise one has 4,250, and a Starter seat has 500 with a daily cap on top.
Decide what happens at zero. Pay-as-you-go at $0.03 a credit with a spend cap, an upgrade, or a queue. Deciding now is cheaper than deciding on the day of a pitch.
The useful output is not a precise forecast. It is knowing whether your team is at 20 percent of its allowance or 200 percent, because those are two entirely different conversations with finance.
Self-studyWhy the vendor pages will not give you a clean answer3 min read▶
Be prepared for genuine ambiguity, and do not let it stall you.
- Official pages contradict each other. Figma's own documentation disagrees on whether its agent is Enterprise-only or available on Professional and Organization Full seats. That is not a reading-comprehension failure on your part.
- Promotions distort the math. A live "unlimited generations" offer makes a credit table temporarily meaningless, and permanently misleading if you plan against it.
- Beta features are often free, then are not. Betas can consume zero credits while they are betas. Budget for the day that ends.
- Exhaustion behaviour is under-documented. What happens when a credit balance hits zero mid-project is not always clearly stated. Test it deliberately on one seat before you find out during a deadline.
The leader move: write down the numbers you are planning against, date them, and put a calendar reminder to re-check in ninety days. Your contract and your tenant's admin console beat any blog post, including this one.
Three decisions, this quarter cards · 8 min live
Everything above is analysis. These three are decisions with your name on them, and each one has a default that will happen anyway if you do not choose.
LiveDecision 1 · Who gets metered generation3 min▶
Credits are allocated per seat and cannot be pooled, so "give everyone a bit" is a real strategy with a real cost: capacity sits idle on people who generate twice a month while your heaviest explorer runs dry in week three.
- Concentrate, do not spread. Put the high allowance where the exploration happens. A brand lead reviewing work does not need a prototyping budget.
- Name a spike path. Pitch weeks and launches are lumpy. Decide in advance whether that is pay-as-you-go with a cap, a temporary upgrade, or a scheduled queue.
- Make consumption visible. If nobody can see the meter, nobody manages it. Ask for a monthly view by seat, the same way you would for any other consumption line.
- Do not fund exploration you will not review. Generating forty options nobody has time to judge is spend, not progress. Capacity to explore should be sized against capacity to decide.
LiveDecision 2 · What you stop paying humans to do, and what you must not3 min▶
The stop list is the easy half and it is genuinely worth money: mechanical resizing, background removal, upscaling, the ninth placement variant, first-pass translation, the shorter version of copy that already exists. If a task has a correct answer and no judgement in it, stop buying human hours for it.
The must-not list is where leaders get this wrong, usually by accident:
- Do not stop paying for the decision. Option selection is the new bottleneck. Under-resourcing it is how you end up with a fast team that ships slowly.
- Do not stop paying for states. Empty, error, loading, long-content and permission cases carry your product's credibility, and generated work reliably omits them.
- Do not stop paying for the junior rung entirely. The production tasks juniors used to learn on are the tasks you just automated. a5 is entirely about what replaces that rung, and doing nothing is a choice with a three-year bill attached.
- Do not stop paying for accountability. Someone senior signs off. That role got more important, not less.
LiveDecision 3 · The training-data default in your own tenant3 min▶
This one has a default that is already in effect, whether or not you have looked at it.
- Figma content training defaults to ON for Starter and Professional teams. On Organization and Enterprise it is off and not toggleable. There are two separate switches - content training and AI features - and they are not the same decision. The policy has been in effect since 15 August 2024, with education and government accounts excluded.
- Canva's training use is governed by your Privacy Settings rather than by your plan tier, so it is an account-level thing somebody has to actually go and look at.
- Rights language is narrower than people assume. Canva's terms say you own your output except output that modifies or incorporates licensed content, and note that outputs may not be unique.
So the decision is not "should we allow AI". It is: for every tool your team touches, who has confirmed the current toggle state, in writing, this quarter? If the answer is nobody, that is tonight's homework and it takes fifteen minutes.
Before session a2 ◐ 45-60 min total
- Audit which tasks actually moved. List your team's ten most common tasks. Mark each one collapsed, unchanged, or harder now. Anything in the "collapsed" column that a human still does by hand is your first process fix.
- Price a month of generation. Ask two designers to log one real week of generating tasks. Multiply out, then compare against the actual credit allowance on their seats - not against the plan name.
- Check your own tenant's training-data toggle. Open the admin settings for every AI tool your team uses, screenshot the current state, and put a date on it. Do this yourself rather than delegating it.
- Write down two tasks you will explicitly not delegate to a generator, with one sentence each on why. Bring them to a2 - they are the seed of your critique standard.
Official sources covered
This session works from vendor documentation as it stood on 8 August 2026, and it names its own uncertainty. Pricing, credit tables and beta status in this space change monthly; where the official pages contradict each other, we say so rather than picking the tidier answer. Certificates and tool training stay with the providers. This page covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · AI made exploration and first drafts cheap. Where did the bottleneck move to?
Making got cheap; judging did not. Forty options is a harder decision than four, and states, edge cases and accountability never moved at all. If your review process was built for three options a week, it is now the constraint.
2 · Your team is on Figma Professional Full seats. What does that tell you about AI capacity?
Credits are per Full seat per month - Starter 500 with a 150 per day cap, Professional 3,000, Organization 3,500, Enterprise 4,250 - and they do not roll over or pool across people. Seats are the cheap, predictable part. Generation is the metered part.
3 · Nobody has touched your Figma team's AI settings. What is the content-training default?
The default depends on your plan, which is exactly why it needs checking rather than assuming. Content training and AI features are also two separate toggles. Canva's training use, by contrast, is governed by Privacy Settings rather than plan tier.