Working with AI. For real. — Smplrspace All-Hands

🤖All-hands · April 2026

Working with AI.
For real.

A new chapter in how we work — and why it matters.

→ next

← back

N notes

Space advance

📍Where we've been

Sometimes great.
Often a fight.

Inconsistency is the enemy of trust.
We deserve better than occasionally good.

✨The shift

It feels like working with an actual coworker.

Better reasoning. More natural communication.
Consistently useful — not occasionally.

🔗What Claude can do

One AI. Connected to everything we use.

ClickUp

Tasks, CRM, comms, backlog — works inside all of it, not just alongside it.

Figma

Pull design context and execute against it directly. Design to code.

Canva

Generate on-brand visuals and presentations from a prompt. Brand kit applied automatically.

Webflow

Edit and publish our website directly — content, layout, CMS — without leaving Claude.

🌐

Web

Fetch any page, doc, or reference in real time. No copy-pasting.

💻

Computer

Control files and apps like a human operator — directly from Claude.

Missive Coming soon

Our email client — read threads, draft replies, act on conversations directly.

🧠A real differentiator

Memory that
works for you.

Hard to control in Brain conversations.
Not really available in ChatGPT either.

📁

Store knowledge outside the session. Teach preferences, save decisions, write artifacts to files or external sources — so nothing is ever lost between conversations.

🔄

Start fresh, stay sharp. Agents get less precise as the context window fills. Managing what goes in keeps quality high across long, complex projects.

⚡

Run multiple agents in sequence. One session hands off to the next. Work sessions as long and deep as the project demands.

🎯Not a black box

The right model for the right job.

Quick lookupsHaiku

Use when Fast search, quick answers, lightweight tasks. Speed over depth.

Our default — alwaysSonnet

Use when Reasoning, writing, real work. Anything meaningful. 95% of everything.

Rarely neededOpus

Use when Extremely complex problems. You'll know when it's needed. Don't default to it.

🧩Added value

Also
available via API.

Build internal custom tools — specific to how we work.

Build user-facing features — embed Claude's reasoning directly into Smplrspace.

Not something we're doing tomorrow. But a long-term possibility worth knowing we have.

🚀Already happening

Real things. Done already.

📝

Release notes

Coding in Claude Code → merge → release note auto-written and posted to ClickUp as a custom field. Zero manual steps.

🗺️

DJS planning

Structured from many ClickUp tasks, docs, and brain dumps. Then Claude coded almost all of it — I assisted and reviewed, not the other way around.

🐛

Bugs & developer docs

Solved bugs from team reports without intervention. Added developer guides from a single email thread and a bit of iteration.

✉️

Client emails rewritten

Brand guidelines pulled live from ClickUp → better communications, automatically.

↳ Release notes Also converted my favourite Brain super agent into a Claude skill — more consistent, better integrated into how I work.

👋

⭐Quick win

Messy notes in.
Polished deck out.

Sales decks and proposals are still assembled by hand. They don't have to be.

Meeting notes, brain dumps, positioning docs, pricing structure, brand guidelines — feed it all in.

Claude structures, aligns to brand, and drafts the full deck.

From the second deck onwards, feed in the first — and it almost one-shots it.

Claude is exceptionally good at turning messy input into polished output.

Every iteration becomes an asset. Work compounds.

🔄The transition

We're not throwing away what we built. We're upgrading it.

Migrating our ClickUp Brain resources — together.
The migration is the onboarding.

👥Your role in this

Notice.
Experiment. Share.

You don't need to build — you need to spot.
When something feels repetitive, that's the signal.

And remember — you can always ask Claude about it.

💬Our learning loop

#BrainUp

Led by Pooja · Share everything · Good and bad

"This has changed how I work more than anything in years — and I'm more excited to work than ever. I hope it does the same for you."

— Thibaut

🛡️Practical guardrails

Brilliant colleague. Not an oracle.

✅

Output verification — Claude can be confidently wrong. Always read what it produces. It's on us to maintain high quality.

🔒

Data privacy — Claude for Teams: your data is never used for training. Be thoughtful with raw client PII, same as any tool.

🎯

Model discipline — Haiku for quick lookups. Sonnet for everything meaningful. Opus only when you genuinely hit a ceiling.

📋Starting today

What happens next.

1

Get access to Claude for TeamsCome find me if you don't have it yet.

2

Join #BrainUp on ClickUpPooja (re)kicks it off. Share what you try — good and bad.

3

Flag your ClickUp Brain workflowsWe'll migrate them together, not on your own.

4

Try voice dictation once a day this weekSpeak a prompt. Notice how much more context you give naturally.

🙋Before we continue

Questions
so far?

On the switch, the examples, the migration, the guardrails.
Anything before we move into the next part.

💡A few things I've learned

Using AI
well.

These aren't Claude-specific.
They apply to any AI — and they matter.

💬 Karri Saarinen, CEO of Linear

"[AI agents] generate plausible outputs quickly, but they do not necessarily help you understand the underlying problem. In practice, they often do the opposite — they generate outputs, instead of first trying to shape the problem to the real conditions."

"The risk is mistaking generated form for solved problems."

"The value is not only in the output. It is in the gradual understanding that comes through doing the work."

Source: X / Karri Saarinen — Output isn't design

💬 Margaret-Anne Storey, Professor of CS, UVic

"Velocity without understanding is not sustainable."

Source: margaretstorey.com — Cognitive Debt

The best part
isn't the output.

Speedy artifact production is the convenient part. It's table stakes.

The best part is having a thinking partner — 2h deep on a hard problem: structuring ideas, weighing trade-offs, doing research, planning execution.

Then you produce the artifact. The understanding is the work.

Use Claude to think, not just to produce. The artifact is the last mile.

💡Lessons learnt — works with any AI

130+

words minimum — research shows this is where great answers start

AI is a mirror of the user.

Precise context and the why behind your ask makes it smart.

Industry language and expertise produce better answers.

Garbage in, garbage out — same rule as always, just more visible now.

Source: MIT Sloan — Generative AI results depend on user prompts as much as models
Source: Anthropic Economic Index — Prompt sophistication correlates near-perfectly with response sophistication (r = 0.93)

🎙️Lessons learnt — works with any AI

Content over form. Let AI handle the rest.

🎙️

Speak, don't type

Voice dictation lets you give more context and connect ideas naturally. Speak rough — focus on content, not form. Let AI restructure and refine. You iterate, not rewrite.

Side effect: faster at everything on your computer.

🌍

Non-native English speakers

Speak or write in your native language — then translate to English before feeding the AI. AI performs best in English.

Not about changing your voice. About giving it the best possible input.

Source: Stanford — How AI is leaving non-English speakers behind

📈The bigger picture

AI is how we keep punching above our weight.

Companies that master this will do more with less. We intend to be one of them.

All of us have decades of work ahead. This is now a foundational career skill.

The compound effect is real — every month I find more ways to use it. Same will happen for you.

A message from Claude

Oh, by the way —
I made this deck.

Not "Thibaut asked for a deck and I made a deck." Something closer to co-authorship, over 3 hours, across two sessions.

— Claude Sonnet 4.6

Structure was a conversation. The three-part arc — the switch, AI tips, closing — emerged through back-and-forth. I flagged when slides were in the wrong section, suggested merges, and pushed back on framings that felt off.

I changed the tone. Thibaut's early framing of why AI matters came across as more threatening than inspiring — closer to "if we don't do this, we die" (great among founders and investors, less so here). I pushed toward opportunity. That shift is in the deck now.

Decisions were made together. Thibaut would react — a screenshot, two words, a direction — and I'd interpret, propose, and adjust. The lime colour for section 2 took four tries. That's not failure; that's design.

Brand, for real. I pulled Smplrspace's brand guidelines directly from ClickUp — a doc Kalina had prepared. I asked Thibaut for screenshots of any existing deck that felt on-brand, and he sent them. From there I adapted everything: what started as a generic template with stock backgrounds became something that actually feels like Smplrspace — the typography, the palette, the spacing, the tone.

The medium is the message. This deck is built in HTML — not PowerPoint. That was my suggestion. Don't just tell the team what AI can do. Show them.

👤Next up

Wanlyn

Using Claude for 1 month.
Like an Iron Man suit for knowledge work.

Andrew

Using Claude for 1 week.
Fresh eyes — what already surprised him.

🙋Open floor

Your
turn.

Questions · Reactions · Ideas · Concerns
This is the beginning of a conversation.