Agentic All Hands · 2nd Attempt · C- Grade.
Lessons below for you to benefit from our innovation failures. Eight from our second agentic all-hands. Most worked. The video gave the room a feeling but missed the message. Cards face up.
Quick scan: what worked, what didn't.
Tap to switch- Everyone in the room participated. No spectators.
- Five people from across the team led mini-TED slots instead of the C-suite.
- The mini-TED speech format kept attention through every minute.
- Cross-team groupings in breakouts surfaced connections that don't show up on any org chart.
- Post-meeting feedback: it landed memorably, with concrete things to think about.
- No red thread tying the hour together. Six great segments, zero through-line.
- Generic breakout instructions. 50 people interpreted them 50 different ways.
- The video gave the room a feeling but missed the message we wanted them to leave with.
- I ran out of tokens on the call. Yes, that happened. Cards face up.
- The structured close felt like a quarterly business review. We dropped it mid-meeting.
All breakout instructions are wrong.
Innovation breakouts only work when the assignment is profile-driven. The room is never one audience.
- Tag every participant by archetype.
- Route a tailored breakout assignment per archetype.
- Let the agents pick on-ramps from prior project data.
- Hold the outcome target constant. Vary the path.
- Read one assignment to fifty people.
- Treat the room as one audience.
- Hope creativity emerges from a generic ask.
- Default to "design-thinking 101" facilitation.
Ours were wrong. 50 people, the same breakout assignment, 50 different interpretations. Some groups went deep on one person. Some filled every field but never talked. One spent ten minutes choosing a photo. Another finished in seven minutes and sat awkwardly.
The room is not one audience.
50 employees · 5 archetypes · 1 set of instructionsSame hour, same outcome target, different on-ramps. The agents already have the data: Notion roles, calendar density, project tags, People DB tier. Generic facilitation fails. Even advanced design-thinking facilitation fails. Profile-driven instructions are the unlock.
FAILOne assignment for fifty people
The breakout brief was a single set of instructions. Builders heard "checklist." Facilitators heard "vibe." Skeptics heard "another waste of time." We needed five briefs, routed by archetype, with one shared outcome.
Claude Design made every employee a top-tier designer overnight.
Claude Design lets any employee operate at the highest level of visual output their company has ever shipped. The bottleneck shifts to ideas.
- Load company branding files for every employee.
- Treat top-tier design as a baseline expectation.
- Spend the saved hours on the message, not the artwork.
- Reuse one design system across decks, dashboards, blogs, social.
- Outsource a deck a skill can produce in ten minutes.
- Pay design tax on a moat that no longer exists.
- Mistake "looks professional" for "says something true."
- Keep the brand kit locked to the design team.
Anthropic's Claude Design release lets any employee load company branding onto their local machine and instantly get the design output that used to cost thousands per slide deck. Templates, color systems, typography, the whole opinionated visual layer. Free.
The practical effect: every employee is now operating at the highest level of visual output your company has ever shipped. The intern can produce a one-pager that looks like the agency's senior designer made it. The CFO can build a board chart that looks like a Bloomberg terminal. The PM can drop a prototype that looks like it shipped last week.
When the design tool gets cheap, the ceiling for "looks professional" rises overnight. Presence, story, and clarity become the only differentiators left.The lesson · Claude Design
For an innovation company, this changes the math on every deliverable. The artifact you ship to a customer is no longer constrained by your in-house design capacity. It's constrained by how well you understand the customer. Which is the only constraint that ever mattered anyway.
FAILTreating design as a moat
We almost hired a contract designer for the all-hands deliverables. Killed the spend the same week. The Claude Design files Charlie shared in Slack did the work in an afternoon. If we'd flinched, we'd have spent five figures on something a skill now produces by default.
The new speech format is the agentically supported mini-TED.
Innovation leaders need 90-second, 4-minute, and 7-minute talks in their tool kit. AI agents and skills carry the technical load. Humans deliver presence and message.
- Match the slot to the message.
- Open with the payoff in the first ten seconds.
- Show a working artifact on screen, not bullets.
- End with a specific ask, owner, and date.
- Default to "I need 30 minutes."
- Open with self-introduction.
- Read a slide deck out loud.
- Drift past your time. Even by 30 seconds.
We borrowed the format from the World Economic Forum. Five mini-TED slots in our hour, none over four minutes, spaced through the agenda and interleaved with breakouts so the room never sat still for long. Three lengths. Three uses. Each one is a mini-TED talk where AI agents and skills carry the technical load: structure stress-test, design system, timing. Every speaker was deeply human because they could focus on presence and message instead of pixels and slides.
A hot signal. One number. One trend. One ask. The fastest format in business.
Human delivers: the signal.
A real idea. Working artifact on screen. Specific ask, specific owner, specific date.
Human delivers: the idea.
A leader sets direction, frames the quarter, or unpacks a case study with arc. The longest slot anyone gets, no exceptions.
Human delivers: the direction.
The 4-minute speech is the hardest format in business. You can't drift. You can't recover. Every word earns its slot. Three things have to land: a clear payoff in the first ten seconds, a working artifact on screen, and a specific ask with an owner and a date at the end. The agents take care of everything else, so the speaker can be fully present.
We're using AI so humans can operate at their most human level. Not the other way around.The philosophy
FAIL 01Timing
We didn't time the speeches. Speakers ran 30% over without realizing. The format only works if you respect the clock. No human moderator wants to cut a colleague off mid-sentence. Let the agent enforce time.
FAIL 02Practice
Speakers practiced once or not at all. Words alone don't carry a 4-minute slot. Presence does. The breath, the pause before the payoff, the eye contact that says "I mean this." Two practice rounds with peer feedback before going live. Non-negotiable.
Cards face up. Or: your team plays the wrong game.
Innovation teams that play poker (cards hidden) duplicate effort and underperform. Teams that play blackjack (cards face up) compound and win.
- Make every workflow, blocker, and bet visible.
- Reward the speaker who admits a hidden silo.
- Ship status updates with the unflattering parts intact.
- Play with the dealer as the only opponent.
- Hold a strategy close until it's "ready to share."
- Polish away the hand you actually have.
- Mistake bluffing for strategy.
- Reward employees who hoard context.
Most teams play poker with their work. Cards held close, hands hidden, every player calculating alone. Blackjack is the opposite. Every card on the table. Every player rooting for everyone else. The dealer is the only opponent.
The two games your team is actually playing.
Poker
Blackjack
Thank you, Chrystina.
Saying "we act siloed and we aren't sharing enough" out loud in front of fifty colleagues is hard. Chrystina did it. She named the hand the team was actually playing, in real time, and put her own cards face up before asking anyone else to. That's the move that makes the deck-of-cards lesson real. The whole post points back to her courage.
FAIL 01Weekly cadence in a daily-ship world
AI agents have compressed work time. Cards face up only works if the cards are actually on the table. The teammate who ships once a week has no hand to play, and the team can't help on a useful timeline. Daily is the new weekly. For some workflows, hourly is the new daily.
FAIL 02Leader cards held close
If the founder plays poker, the team plays poker. We've held strategy decks until they were "ready to share" more than once this year. Each time, we lost weeks of compounding. Cards face up starts at the top, and starts daily.
Tokens are the most honest metric a team has.
Token spend reveals how AI-native a team really is. Make it face-up and watch behavior change.
- Show weekly burn by tier.
- Celebrate heavy builders.
- Coach light users with one workflow at a time.
- Tie spend to output, not just cost.
- Treat spend as a private cost-center number.
- Reward "low usage" as if it were efficiency.
- Penalize the heavy builder before you understand their output.
- Hide the spread from the team you need to align.
Thank you, Jared.
Putting actual weekly token spend on a screen in front of the whole company is uncomfortable. Jared did it. Real numbers, real spread, real surprises. The room learned more about how each of us actually works in 90 seconds than in any quarterly review.
Weekly token spend, face up.
Want the token counter we used? I'll send you the code.
The same skill Jared ran in front of fifty people. Drop it on your team and you'll have face-up token data by next Friday. Free. DM me and I'll send the code.
FAILI ran out of tokens on the call
Out loud. In front of the team. While trying to send a transcript to Brian. Cards face up means I have to admit that, too. Visibility cuts both ways. Someone on the team should always be willing to show their hand first. This time it was me.
Feeling carried the room. The message got lost.
The hardest job a leader has is empowering every speaker to say bold things. The red thread isn't a topic. It's permission.
- Brief speakers on the feeling each block should leave behind.
- Connect every block to one message the room can repeat at lunch.
- Earn each speaker the permission to be brave.
- Make a video to "give the room a feeling" and skip the message.
- Confuse engagement for retention.
- Leave speakers to find their own courage cold.
I shot a video for the all-hands. The video gave the room a feeling. People leaned in. But the message I wanted them to take home didn't land. Six segments scattered into six memories. We needed one sentence the room could repeat at lunch. We didn't pick one.
What they felt.
Energy. Surprise. Pride at being asked to show up differently.
Easy to land. Easy to over-rely on.
What we wanted them to remember.
One sentence the room could repeat at lunch. We didn't pick one.
Hard to design. The whole job.
A great segment without a theme is a great song that nobody hums afterward.Field note
The red thread isn't "innovation." It's permission. Permission for every speaker to say something true and uncomfortable, knowing the room will catch them. Each speaker had earned the room only because of the segment immediately before them. Pull one out and the rest get harder.
FAILFeeling without message
I gave the room a feeling and forgot the takeaway. Engagement went up, retention went down. Without one shared message, we had a great hour and no shared next step. That's theater, not leadership.
The About Us page is the personal innovation surface almost every company is wasting.
An innovation company's About Us isn't a static team grid. It's a network of personal surfaces, each tied to a topic that employee cares deeply about. OpenAI and Google scrape that surface daily. Most companies aren't using it themselves.
- Give every employee their own page, not a tile in a grid.
- Let each employee claim one topic they care deeply about.
- Let agents update the page from real work signals.
- Treat About Us as a recruiting and sales asset, not marketing fluff.
- Ship the generic SaaS team grid every other company has.
- Lock the page so only marketing can edit.
- Confuse a prototype with production.
- Wait for the page to be "perfect" before going live.
Two messages worth carrying out of this hour. One: the whole room worked on our About Us page together during the call, collectively improving each other's profiles. Fifty people. Twenty minutes. One hour of agent editing afterward. What came out of it is the most personable representation of an innovation company I've ever seen.
Two: the digital exhaust your company emits every day, the same fumes Google and OpenAI scrape to fuel their models, can fuel your own team. Each employee already has a topic they care deeply about. Each topic is already producing exhaust on the internet. The unlock is collecting that exhaust into a personal surface that points back to your company. Most companies leave it on the floor.
Coming: the new About Us.
Ships in 2 weeksFAILMistaking prototype for production
The 20-minute prototype was electric. The instinct was to ship it the same week. We didn't, and we shouldn't have. Production is about preserving the prototype's energy without losing the rigor: SEO, AEO, public traffic, and the awkward fact that some employees aren't sure how visible they want to be on the open internet. Two weeks. Not two days.
The whole hour was planned in one Claude Code session.
No project managers. No designers. No speechwriters. No external agency. One five-minute direction call, then skills did the rest. Each speaker spent their time on presence, not pixels.
- Plan the whole event in one agentic session.
- Hand each speaker a skill to draft, design, and time their slot.
- Reserve human time for practice and presence.
- Default to "skills do this, humans deliver that."
- Run a six-week planning track for a one-hour event.
- Hire a designer for what a skill produces in minutes.
- Use a project manager to chase status that's already visible.
- Confuse coordination overhead with progress.
One five-minute call gave direction. Everything after that ran inside a single Claude Code session. Skills drafted the agenda, briefed each speaker, designed the slides, timed each slot, and routed conversations to the right teammates and the right agents on the right computers. Zero coordination calls. Zero status meetings. Zero external help.
An all-hands is one of the most expensive meetings a company runs. Pull 50 people out of customer work for an hour, and that's 50 person-hours gone before anyone even prepares. Recent meeting-cost research ranks all-hands and town halls among the top categories of unproductive meeting time, and Asana's Anatomy of Work tracks the bigger pattern: knowledge workers lose the majority of their day to coordination rather than the work itself. The all-hands is where that pattern is most visible. Most teams layer dozens of hours of prep on top of the meeting itself, across people who never even speak.
Each speaker spent their saved hours on the only thing only they could do: practicing presence, honing the message, calling on a teammate or skill when they wanted polish. That's the trade. The agents take the coordination cost. The humans take the room.
The agents take the coordination cost. The humans take the room.The trade · agentic planning
FAILCutting the support layer too far
The day didn't go entirely smoothly. The speakers carried a small unease, and the room felt it. That's the cost of removing every human support role at once. Next time we'll keep one human in the loop: a 30-minute peer rehearsal the day before, no agency required, just a teammate with sharp ears. Agentic planning isn't "no humans." It's "humans on the parts that matter."
Field FAQ
Why are most breakout instructions wrong?
Most breakout instructions are written as a single set of prompts read to the entire room. But a 50-person room contains multiple archetypes: builders, facilitators, veterans, skeptics, new joiners. Profile-driven instructions tag each participant by archetype and route a tailored prompt with the same outcome target.
What is a mini-TED format for innovation all-hands?
Three time slots: 90 seconds, 4 minutes, and 7 minutes. 90 seconds for a hot signal. 4 minutes for an idea with a working artifact. 7 minutes for direction, table-setting, or a case study. Each is supported by AI agents and skills so the speaker focuses on presence and message instead of pixels.
What is Claude Design and why does it change innovation work?
Claude Design is Anthropic's release that lets any employee load company branding onto a local machine and produce design output that used to cost thousands per slide deck. Every employee now operates at the highest level of visual output the company has ever shipped. The bottleneck shifts from production to ideas.
What is "cards face up" as a team operating model?
A team plays poker when each person works alone with their hand hidden. A team plays blackjack when every workflow, blocker, and bet is visible. Cards-face-up teams ship faster because nothing is solved twice. With AI agents compressing work time, daily-ship cadence is the new minimum for face-up to actually work.
Why is token spend the most honest metric of an AI-native team?
Token spend reveals which team members actually build with AI versus which talk about it. A ~100x spread between heavy builders and light users on the same team means different operating models, not different tools.
What is the difference between a meeting's feeling and its message?
Feeling is what the room walks out with emotionally. Message is what they can repeat at lunch. Feeling without message is theater. Innovation leaders design both, then connect every segment with a single thread of permission so every speaker can land bold truths without flinching.
What does an innovation company's About Us page look like?
Not a static team grid. A network of personal surfaces, each tied to one topic that employee cares deeply about. The same digital exhaust Google and OpenAI scrape every day can fuel your own team's discoverability. Most companies leave that surface unused.
How do you plan an all-hands agentically?
One five-minute direction call to align the leadership team. Then a single Claude Code session with trained skills handles agenda design, speaker briefs, slide design, timing, and coordination. No project managers, designers, or external agencies required. Speakers spend the saved time on practice and presence.
Take any of these off our hands.
Each one ships free. Reach out on LinkedIn or via the Collaboration.Ai contact page and I'll send the files, the markdown, or the code.
The token counter code.
Drop our token counter app on your team and you'll have face-up token data by next Friday. Same skill Jared ran live in the all-hands.
The all-hands playbook.
Step-by-step for running your own innovation all-hands using this format. Agenda template, breakout instructions, NWCU speaker brief, the works.
Claude Design setup.
How we got every employee operating at our highest design level overnight. Our brand files, our setup, our gotchas. No agency required.
The 4-minute timer skill.
A skill that times speeches inside Google Meet and Zoom. Pings the speaker at 60 and 80 seconds so the agent enforces the clock, not a human.
If you want the original setup that got us here, read the first post: Your All-Hands & Town Hall Needs an Innovation Refresh. The post you just read assumes you got that far.
Brandon Klein is co-founder of Collaboration.Ai. We build agentic innovation infrastructure. We also run our own all-hands like the customers we want to have.
