Loading...
Case study from our own kitchen
ThinkagainOS: our company runs on its own operating system
We believe the most interesting organisations of the coming years will be AI-first. That goes beyond 'doing something with AI': you design your company so AI knows your full context and works along everywhere. ThinkagainOS is our version of that: one workspace of our own for all our work, with a company brain on top. On this page we explain how it works, what it gets us, and what you can take from it.
Everything you see below was rebuilt with fictional sample data, but this is exactly what our real work looks like.
20+
modules in one workspace
1
brain across all our work
100%
built ourselves, together with AI
One system instead of fifteen tools
ThinkagainOS is our own dashboard where we do everything: manage the website, follow up with clients, plan social media, write documents, build presentations and talk to each other. That used to take a stack of separate subscriptions that didn't know about each other; here everything runs in one environment on one database. It sounds ambitious. Yet it started very small, more on that in a minute.
Articles
our CMS
Social media
plan & publish
Pipeline
our CRM
Contacts & companies
everything logged
Chat
AI with our brain
Messages
our Slack
linked to the CRM
To-do
our kanban board
Projects
with templates
Calendar
linked to deals
Bookings
our Calendly
Documents
our Word
Spreadsheets
our Excel
Presentations
our PowerPoint
Files
our Drive
Knowledge base
the brain itself
Reading list
shared sources
Video calls
our Meet
thinkagain.nl/dashboard/kanban
Fictional sample dataDashboard › To DoNewSearch...⌘KAI⌘.
To Do
Manage tasks for the team.
BoardTable
StandupNew taskBacklog3
TASK-131
Prepare keynote: AI in retail
Think again
Presentations21 JunEM
TASK-134
Schedule June newsletter
Think again
ContentPM
TASK-128
Finish SEA research
Think again
SalesPM
New task
In progress2
TASK-142
Proposal phase 2 - Studio Helder
Think again
Sales15 JunPM
TASK-139
Refresh workshop materials
Think again
TrainingEM
New task
Done2
TASK-118
Publish ThinkagainOS case page
Think again
WebsiteEM
TASK-121
Finish Q2 report
Think again
AdminPM
New task
It started with a CMS and a to-do list
We never set out to build a 'platform'. We had a website we wanted to update faster, so we built our own CMS. Then we needed somewhere to put our tasks, so a to-do board followed. Then a place for contacts and deals. Every time something in our daily work felt clumsy, we built the missing piece, at our own pace and shaped around the way we actually work.
The idea of a second brain isn't new. Andrej Karpathy described how he kept all of his knowledge in Obsidian for years, and that inspired us. With one big question on top: what happens when you do this with an entire company instead of one person, and let the brain live inside the system where the work happens?
Our own CMS
Updating the website without friction. The first brick, no master plan.
To-do and CRM
Tasks, contacts, companies and a sales pipeline. Our daily work got a single address.
The full workspace
Documents, spreadsheets, presentations, mail, messages, calendar. The standalone tools went out one by one.
Only then: the AI layer
Once a substantial part of our work and processes lived inside, we laid a brain on top. That order is the whole point.
“If you properly log everything you do, from to-dos and documents to briefings, evaluations, strategies and agreements, you're essentially building a complete knowledge base of how you run your company.”
Erik, internal vision note, 20 March 2026
The company brain
The company brain is all of our work, made searchable. Every note, document, deal and agreement can become part of a single memory that we query in plain language, and that backs up its answers with sources.
Four layers, one system
Everything AI can help with falls roughly into four layers. This is what they look like for us:
One workspace, one databaseWhere the knowledge is bornWhere the context livesWhere it comes togetherWhere the value sits
Where the work is made
Documents · Spreadsheets · Presentations
Where the conversations live
Messages · Mail · Video calls
Where the memory lives
Knowledge base · Files · Reading list
Where the business runs
Pipeline · Contacts · Projects · Calendar
In many organisations these are four separate systems, each with its own integrations and its own permissions, and AI has to learn to reach them one by one. Here, all four layers live in the same database. The brain reaches everything just as easily, and permissions apply everywhere automatically.
Working is capturing
The proposal we were writing anyway, the notes from a client call, the deal in the pipeline: that is the knowledge the brain draws from. Writing documentation as a separate chore never comes into it.
Ask in plain language
"What did we agree on back then?" The brain searches everything it's allowed to see and answers with sources, so you can verify it.
The brain looks back, too
It writes knowledge pages from our work on its own, guards their quality, and flags where knowledge is missing or going stale.
thinkagain.nl/dashboard/chat
Fictional sample dataDashboard › ChatNewSearch...⌘KAI⌘.
What did we agree with Studio Helder about phase 2?
Brain retrievedBrain retrieved
Sources (3)
- [1]Proposal Studio Helder - phase 2Document
- [2]Re: audit planningMail
- [3]Studio Helder - phase 2Deal
The proposal of 14 April describes phase 2 as three workshops plus an AI search audit [1]. In the email thread of 28 April the audit was moved to September at Lisa's request [2]. The deal is marked 'agreed' and the first workshop is scheduled for 3 October [3].
Ask a question or give an instruction...
Sonnet 4.6Workspace knowledgeWeb searchThinking
AI can make mistakes. Check important information.
thinkagain.nl/dashboard/kennisbank
Fictional sample dataDashboard › Knowledge baseNewSearch...⌘KAI⌘.
Sales process: from first contact to proposal
Updated 2 hours ago · v7 · 18 sources ·Quality 87
Findings from the quality check
Not (fully) covered by sources:
- Three recent deals deviate from the standard price structure, but no document describes the pricing strategy.
- New client onboarding has been asked about 14 times, but only one short note exists.
This page describes how a first conversation moves towards a proposal: who does what, which documents are involved and which lead times are realistic. Compiled from 18 sources in the workspace…
Under the hood: how the brain works
Between your question and the answer sits a chain of five steps, completed in a few seconds. The approach is called RAG (retrieval-augmented generation), and you can follow it just fine without a technical background.
1.Capture
Everything we make already lives in one database: documents, deals, mails, notes and tasks. There is no separate 'knowledge step', because working is capturing.
2.Index
Every text automatically gets a meaning-fingerprint (an embedding). That lets the brain search by meaning instead of keywords: a question about 'price agreements' also finds documents about rates and quotes.
3.Retrieve
Ask a question and the brain searches everything it's allowed to see for the most relevant fragments, across thousands of pieces of text, in seconds.
4.Rerank and filter
A second model holds those fragments against your question once more (reranking) and sorts by true relevance. Your permissions apply at the same time: what you can't see, the brain won't see for you either.
5.Answer with sources
The language model writes the answer based solely on those fragments and cites a source per claim. If it finds nothing, it says so instead of making something up.
On top of that, the brain continuously writes knowledge pages per topic on its own, like the sales process example above. That's the step from search to synthesis: loose fragments become living summaries that grow with the work, including a quality check that flags missing sources.
Under the surface this runs on our own database (Postgres with pgvector for the embeddings) and language models from Anthropic and OpenAI. The special part is the combination: everything in one place, with permissions, inside the system where the work already happens.
We decide what the brain knows
Not everything belongs in a shared memory. Per document, folder or module we decide whether it joins the brain, and who gets to see it. Personal notes stay personal, financial documents stay with the directors, and the brain respects those boundaries in every answer.
thinkagain.nl/dashboard/notities
Fictional sample dataDashboard › DocumentsNewSearch...⌘KAI⌘.
Strategy 2026
SavedFeeds the knowledge baseTeamFinance Q2
SavedExcludedDirectors onlyPersonal notes
SavedExcludedOnly meDocument excluded from the knowledge base
The honest truth: a brain is only as good as what you feed it
This is where most second brains die. The system lives next to the work, so keeping it up to date feels like extra work, so it quietly fades away. That's exactly why ThinkagainOS became a complete workspace rather than a standalone knowledge base.
Because we do all our work inside the system, from mail and documents to deals and conversations, the brain feeds itself. Capturing knowledge is no longer a separate chore; it's a by-product of simply doing your job.
The more we work in it, the smarter the brain gets. The smarter the brain, the more it takes off our plate. The more it takes over, the more time is left for the work that actually matters.
The biggest advantage: it's never finished
ThinkagainOS isn't a product that was ever 'delivered'. It grows along with how we work. If something's missing, we build it. With AI as a building partner that takes hours instead of weeks: sketched in a morning session, live for the whole team that afternoon.
That fundamentally changes your relationship with software: the tool adapts to us, rather than the other way around.
What this says about the future of work
There's a difference between applying AI and building AI-first. Applying AI is adding a loose tool for a loose task. Building AI-first means designing your processes, knowledge and systems so AI has context everywhere. And context is exactly what lifts AI from impressive to indispensable: a model that knows your clients, agreements and history is in a different league than an empty chat box.
For us, work is shifting from executing to directing. Every project makes the next one easier, because the system learns along. The most interesting question for the coming years, as far as we're concerned: where does your company's knowledge live, and can AI reach it?
“In the end you're left with just one role: the conductor of your own company.”
From the same note, 20 March
Want to think this through for yourself? Start here
You don't need to build your own software to put this to work. The thinking steps are the same for every company.
1
Start with your daily work
Pick the process you touch every single day (for us that was the website and the task list) and make sure it lives in one place. The AI comes into the picture after that.
2
Capture your processes before adding AI
A brain without knowledge is an empty box, and AI on top of a messy foundation mostly amplifies the mess. The magic is in the order.
3
Choose a system you genuinely work in
Notion, Obsidian or something of your own: if capturing is extra work, nobody keeps it up. It has to be the place where the work already happens.
Frequently asked questions
Isn't this just ChatGPT with your documents attached?
No, and the difference comes down to three things. One: the brain lives inside the system where we work, so it's always up to date without anyone uploading or maintaining anything. Two: it knows the connections. A deal is linked to a company, to mails, documents and tasks, and the brain reasons over that whole. Three: it respects permissions and backs every answer with sources, so you can verify it.
What if the AI makes something up?
The chain above keeps that risk small: the model only answers based on retrieved fragments and cites a source per claim. If it finds nothing, it says 'that hasn't been captured' instead of guessing. And since every source is one click away, you verify important answers in seconds.
What about privacy and sensitive information?
Per document or folder, we decide whether something joins the brain. And the workspace permissions apply to the brain too: it only retrieves what the person asking is allowed to see. Financial documents stay with the directors and personal notes with their owner, in the answers as well.
Do you need to know how to code to pull this off?
We build ThinkagainOS ourselves, with AI as a building partner. That's what makes it feasible for a small team. But the mental model matters more than the technology: one place where your work lives, capturing processes before adding AI, and control over what the brain may know. You can take those steps with existing tools like Notion or Obsidian, too.
Want to talk this through?
We build ThinkagainOS first and foremost for ourselves. But few things are more fun than thinking out loud about what this could look like for your organisation. Send us a message and we'll happily share what we've learned.