Outcomes
What the team leaves able to do
We judge the session by what attendees can do on their own the next morning, not by how impressive the demo was.
Make it
Turn a plain-language request, typed or spoken, into an on-brand piece with the skill.
Direct it
Give art direction in their own words: what's off-brand, what to change, which format they need.
Save it
End each working session by asking Claude to document what they learned, so the skill gets smarter.
Own it
Know who updates the skill, how changes reach Hyperakt, and when version 2 arrives.
Who's in the room
Four roles, named in advance
1–2 people
Skill owners
The brand owners who decide what goes into the skill. Usually the comms or creative lead. They're the only ones who update it.
3–6 people
Makers
Anyone who makes graphics, decks, social posts or web pages today. Each one brings a real task.
Pre-work only
Client IT admin
Whoever administers their Claude account. We need them the week before, not in the session.
Hyperakt
Lead and second
The lead runs the session and art-directs. The second fixes setup snags one-on-one and takes notes for the pilot log, including who lights up during hands-on.
Optional
Outside vendors
The client's web, print or merch agency. Check their plan and font license in the pre-work, the same as the client's. They need prototyping and handoff, not brand basics.
Before the session · one week out
Settle setup before anyone walks in
In the pilot, the hands-on part hit permissions, installs and fonts that nobody had checked. This list makes sure the session is about making, not troubleshooting.
Client admin checklist
- Confirm the plan (Team or Enterprise) and who the account owner is.
- Turn off model training for the whole organization: Settings › Privacy › "Help improve our AI models."
- Confirm the license covers AI features for every attendee's seat.
- Install the Claude desktop app for every attendee. Cowork and Code need it; the browser isn't enough.
- Upload and turn on the skill for the organization. Check that each attendee can see it.
- Decide who gets it. On Enterprise, the admin can limit the skill to a group (for example, comms only). Set it up now, not after the session.
- Vendors in the room? Confirm their own Claude plan, and get the client's written OK for them to use the brand fonts (or send a version of the skill without the fonts).
- Install the brand fonts on attendee machines, only the weights the brand actually uses.
- Check domain limits. Make sure Claude can read the client's website and partner sites (for logos and content).
- Give Claude a photo library: a dedicated, tagged folder or service it can reach.
Attendee intake
- One recurring task they'd love to hand off (a weekly social post, a slide type, an event graphic).
- Their role and the tools they make things in today.
- One idea they've wanted to build but couldn't justify the cost of.
Hyperakt prep
- Check that it's installed and opened, not just sent. Two days out, ask each attendee to run one starter prompt and reply with a screenshot.
- Dry-run every intake task with the skill. Fix what fails in the skill, not in the room.
- Pre-run a fallback for each category likely to come up (web page, merch, display banners, deck). Long builds take 5–20 minutes, so keep a finished, unedited result ready to show while the live one runs.
- Pick the live build: the intake idea with the biggest "we were quoted how much?" story.
- Agree on skill owners with the client lead and plan how updates will work (see Ownership).
- Prepare an honest answer on AI disclosure for their sector (see Tough questions).
The session · 90 minutes
Run of show
Hands-on starts at minute 25 and takes the biggest block. Everything else supports it. No one on our side talks for more than eight minutes without asking the room a question.
Their work, read back
Open with the tasks they sent. Read a few aloud, so everyone hears that today is about their work. Then give the one-line rule for which Claude tool to use for what (see Which tool). Before anyone touches it, say: "Nothing you try today can change anyone else's skill. Your experiments stay in your own Claude."
Lead's moveName each attendee's task out loud. Skip the brand guide tour: the skill already knows it. The fear of breaking the shared skill is what stops people from exploring, so settle it first.
Done whenEveryone knows which task they'll make today and where they'll make it.
Live build from their idea
Ask one attendee to describe the thing they've wanted to build. Record them talking, drop the recording into Claude with "use the design skill to mock this up," and let it run while you keep going. Reveal it at the end of the block.
Lead's moveStart it early and come back to it. The wait is part of the lesson: talk to it like a person, then step away. If someone mentions a real task at any point ("I had to make banners yesterday"), switch to it. A task they bring beats anything we prepared.
Done whenThe room sees their own idea, built from their own components, in under 15 minutes.
Everyone makes their task
Each attendee builds the task they brought. The lead moves around the room, pushes people to do at least two rounds of changes, and shares good results on screen. The second deals with any setup snag one-on-one, so the room keeps moving.
Lead's moveCoach the asking, not the clicking. "Tell it what feels off" beats "click here." If a group joins from one conference room, pick a driver: one of them types on their own machine, shared on screen, while the room tells them what to ask.
Done whenEvery attendee has one piece they'd actually use this week.
Art direction, out loud
Pick two pieces from the hands-on block and critique them live. Say what's off-brand and why. Ask for three color options. Ask for a different format: a PNG with a transparent background, a vector file, a slide.
Lead's moveModel the judgment. The skill knows the rules; the team's eye decides what ships.
Done whenEach attendee has given at least one round of direction in their own words.
Ownership and the loop
Name the skill owners. Show how an update works. Teach the habit: end every working session with "document what we did and learned." Explain the monthly lessons document that comes back to Hyperakt.
Lead's moveSettle who can update the skill before the session (see Ownership), so this block is a decision already made, not a debate. Use the scripted answer, word for word.
Done whenOwners are named and everyone can repeat the loop.
What else is possible
A short show-and-tell, kept separate so it doesn't take over the session: decks driven from Claude, interactive pages published on their own subdomain, project memory, small internal tools. This is where the next engagement starts.
Lead's moveKeep it to three examples, each tied to something they said earlier. Before closing, set the Day 14 date out loud.
Done whenThey name one thing they want next, and the second has noted who lit up for a follow-up conversation.
When we only get 45 minutes
0:00 · 5 min
Their work
Read back the tasks, give the one-line tool rule, and say the "your experiments can't change anyone else's skill" line.
0:05 · 10 min
Live build
Start their idea, or the pre-run fallback for anything over 10 minutes. Reveal it before minute 15.
0:15 · 20 min
Hands-on
Starts by minute 20 at the latest. Everyone, or one driver per room, makes their task with two rounds of direction.
0:35 · 10 min
Own and next
The scripted ownership answer, the document-what-we-did habit and the Day 14 date. Cut the show-and-tell before you cut this.
Their idea, built in front of them, beats any demo we prepare.
Which tool
One line for each tool
Too many options confused people in the pilot. The skill works in all three, so hand them this rule and move on. They don't need Claude Design.
Chat
For quick, single piecesA social graphic, a slide image, a caption, a resized version. Fastest way to get going.
Cowork
For working with your filesDecks, documents and folders on your computer. Point it at your template or your notes and let it build.
Code
For pages and anything you publishWeb pages, interactive pieces, motion and anything kept in a repo. You can keep adding requests while it works.
How we talk about it
Show it, don't promise it
Every big claim sets up a later letdown, and the output still has small flaws a designer would catch. Let the work on screen make the case.
Say
- "You're the creative director." It starts you on brand; you decide what ships.
- "It replaces the tedious tasks, not the people." The resizes and the logo swaps, so there's time for ideas.
- "Let's try it." Run the request instead of describing what it can do.
- "It works alongside Canva and Figma, and can write to both."
- "Expect a few rounds of direction." A day's job might take an hour, not one prompt.
Avoid
- "It can do anything." Name what you've seen it do.
- Percentages like "95% of the way there." Nobody can check them, and someone will remember them.
- "This will replace Canva" (or Figma, or a vendor). Especially with that vendor in the room.
- Formats we haven't tested with this skill, such as EPS or Illustrator files. Say "we'll confirm."
- "It does everything the Figma does and more." Say what's in the digital guide instead.
Tough questions
Answers we bring with us
These come up in every room. Mission-driven teams ask them because they care, so answer plainly and don't get defensive.
Will Anthropic train its models on our brand?
Not if the organization's admin turns off "Help improve our AI models" in Settings › Privacy. On Team and Enterprise plans that setting sits with the admin, which is why it's on the pre-work checklist. Confirm it's off before the session.
Doesn't this expose our brand?
Anything already on your website can be scraped by anyone. Brand assets you publish anyway are safe to use here. Keep private material out: donor records, personnel matters, anything you wouldn't post.
Do we have to disclose that we use AI?
Our position: the skill does what your team already does, faster, using elements already in your brand. We don't generate photos, video or new brand elements, and a person decides what ships. For public-facing work, add that line to your AI policy and have your counsel review it. We're designers, not lawyers, so we say so.
Will this replace our designers, illustrators or video makers?
No. Original art, new brand elements, storytelling and video stay with people. The skill takes the production off their plates (the resizes, the event graphics, the deck cleanup) so they have time for that work.
Will it change a partner's logo or our colors?
It works from official files and the brand's rules, and you still check every piece before it ships. If something's off, say so. That correction is how the skill learns.
Can anyone on staff change the skill?
It depends on the plan and how the admin sets it up. Most teams want one or two owners. See Ownership.
Will my experiments change the skill for everyone else?
No. What you try stays in your own Claude, in its memory if you have memory on. The shared skill only changes when the owners install a new version. Say this early: it's what frees people to explore.
Can we add work we made outside the skill?
Yes, through the loop. Send the pieces you love to the skill owner, who passes them to Hyperakt, and they go into the next version for everyone. During our project, that's part of the scope. After it, it's what the Keep It Current retainer covers.
Which model should we use?
The one the skill was built and tested with, named on the guide's skill page (Opus 5.5 at medium effort as of October 2026). Models change often. We re-test the skill when they do, so we'll tell you when to switch.
What will this cost us in usage?
It runs on your own Claude plan. Design work with several rounds of direction uses more than writing, and long memory adds to every request. Give the heaviest makers the bigger seats, and keep everyone else on the standard ones.
Will this replace Canva (or Figma)?
It works alongside them. It can write designs into Canva and Figma, so keep the tools your team likes and drop only the ones you no longer need. Don't promise a replacement, especially with a vendor in the room.
Can it give us Illustrator, EPS or SVG files?
PNG, PDF and SVG, yes. For anything else, check with this skill before the session, or say "tell us the format and we'll confirm." Don't promise a format live that we haven't tested.
Who hosts the digital guide, and can we have it?
Hyperakt hosts it for now. How it's handed over isn't decided yet (a zip, hosting as part of the retainer, or a move to the client's own subdomain). Say "we'll work that out with you" and log the question. Don't improvise a policy.
Ownership
Who keeps the skill current
Updating a skill is easy: "bake what we just did into the skill." That's why only owners should do it. Decide the setup before the session.
Team plan
Only account owners can update an organization skill
Owners have full admin rights, so this only works if they're the people you'd trust with the skill anyway. If an outside IT firm is the owner, they upload each new version (a .zip) that Hyperakt sends.
Workaround
Keep the skill in a GitHub repo
Package the skill as a plugin in a GitHub repo and have the owner connect the repo once. From then on, anyone with write access to the repo can update the skill. A GitHub organization is free and becomes the home for their pages and tools too.
Lock it down
Keep everyone else out
The admin can turn off user-created skills, or require a review before anything is published. Ask for this in the pre-work.
The learning loop
- MakeThe team uses the skill on real work.
- Document"Document what we did and learned." Claude saves a memory note.
- CollectEach month, owners ask Claude to combine the notes into one lessons doc.
- UpdateHyperakt folds the lessons into the skill.
- ShipThe new version goes back to the owners. Repeat.
What we say, word for word
"Who changes the skill, and how?"
"Nobody edits their own copy, because the next version replaces it and the edits are lost. If something's wrong or missing, or you made something you love, tell [owner]. It goes to us and into the next version for everyone. During our project, we make those updates. After that, they come through the Keep It Current retainer. Your first new version is on [Day 14 date]."
Don't say "keep teaching it" or "once our scope ends, the skill stops changing." The first suggests people edit their own copies; the second ends the relationship we want to keep.
After the session
Keep the momentum
Day 1
Recap and to-dos
Send key takeaways and the client's list, with an owner and date on each item. Thank the person who brought the live-build idea.
Day 14
Version 2
Collect the first lessons document, update the skill and send it back. The team sees their feedback in the tool.
Day 30
Measure
Count the pieces made, the hours saved against the old way, and the work they stopped outsourcing. That becomes the case study and the case for what comes next.
Pilot notes
v0.2 · Area Foundation + Parallel · Oct 29, 2026
Our first session with a client's outside vendor (Parallel, their web agency), 45 minutes, with SAAF and Parallel in the room. Strong content, but it ran as a demo: nobody outside Hyperakt built anything.
What worked
- Starting the long website build before the call, and showing a finished, unedited one while it ran.
- "You're the creative director. You're not starting from a blank canvas."
- Leaving content strategy to the vendor as their work. The skill sped up their process instead of competing with it.
- Answering "will my experiments change the shared skill?" with a clear no. The attendee who asked said she'd now explore more.
- A concrete story: a change to the logo's weight, pushed across hundreds of uses with Claude.
- Opening ongoing updates as a service when the client asked to add work made outside the skill.
What we changed
- No one got hands-on in 45 minutes. There's now a 45-minute version with hands-on by minute 20, and a driver for conference-room groups.
- The vendor had received the skill but never opened it. Pre-work now asks for a screenshot of one starter prompt.
- An attendee mentioned a real task (a set of display banners) and we stayed on our own demo. The lead now switches to any real task that comes up.
- About 11 minutes of talking before the first question. The eight-minute rule now applies here too.
- The answer on who changes the skill contradicted itself. It's now scripted.
- Claims like "it can do anything" and "it'll replace Canva" went too far. There's now a language guide.
- Limiting access on Enterprise, vendor font licenses, usage costs and guide hosting came up without answers. They're now pre-work or prepared answers.
- It ended with "go play with it" again. The closing block now sets the Day 14 date out loud.
v0.1 · Brooklyn Org · Oct 8, 2026
We ran the first session with no structure, on purpose, to see what worked. This playbook is what we learned.
What worked
- Building an attendee's idea live from a recording of them describing it. The strongest moment of the session.
- Tying each demo to their own costs: an interactive map they'd been quoted six figures for, an explainer video that took a month.
- Work made from their own content (their prize winners, their grantees) rather than generic samples.
- Answering privacy and ethics questions plainly, in line with their values.
- Critiquing a slide live, which modeled the judgment they need.
What we changed
- Plan, admin and permission questions came up live. They're now pre-work.
- Hands-on started more than an hour in, and setup snags stalled it. It now starts at minute 25 with a second person handling snags.
- The brand guide tour ate time. It's cut, because the skill already knows the guide.
- The Hyperakt tools tour took over the session. It's now a fenced 10-minute closer.
- Chat, Cowork, Code and Claude Design confused people. There's now one line for each.
- The AI disclosure question needed a better answer. It's now prepared.
- "Go play with it" was too loose. Homework is now the Day 14 lessons document.
Changelog · v0.2 (Oct 30, 2026): from the Area Foundation + Parallel session. Added the 45-minute version, a vendor role, pre-work install check and fallbacks, the "your experiments can't change the skill" opener, a scripted ownership answer, a language guide and seven prepared answers. · v0.1 (Oct 8, 2026): first playbook, written from the Brooklyn Org pilot.