13 Sept 2026 · ai, automation, docs
Before you get to shareholder value
Everyone’s shopping for the next Agent, harness, latest model or flashy hype thing. Most companies haven’t sorted out the two boring realities that actually gate whether any of it works scalably. Unfortunately you can’t oneshot-buy either of these & it involves thinking with the brain. As always, this info will probably be relevant for the next ~3 months max & then I’ll dump it because it’s stale.
Your company knowledge has to be AI-ready
Documentation. I mean documentation. And processes + culture & all that good stuff. Documentation is now infrastructure. Documentation is now code.
TL;DR: Digitalize all things. Every info with value needs to live as text. You need to build a maintainable map and processes.
Your AI implementation at scale will flop if your organizational knowledge infrastructure is f*cked.
Ok, but why won’t I see value from my cool latest agentic tool & widespread chatGPT adoption if I’m not locked in on documentation?
Easy: human workers compensate for bad information architecture with context, memory, social networks, asking other humans questions, and of course, hallucinating khm, I mean just figuring it out . Agents don’t have those escape hatches in the same way.
=> AI raises the minimum standard for organizational infrastructure.
It’s an amplifier. Documentation was a problem before as well. Now it’s just 2000x the size. Humans red a doc perhaps 2x per month. It was a bit stale and no one cared. Now your entire company queries the page literally 4000x, on the background, which leads to 4000 slightly/awfully misaligned pieces of advice or work.
OK, what to do?
- Exists. Your documentation has to exist first.
Documentation can happen for different reasons: knowledge, legal, fun, accidental, working documents, scribbly notes, maybe we’ll need it in the future, logs, tech docs etc. Not all of these have the same value, but make sure that everything that does have value, exists in the physical realm (& I don’t just mean in a few synapses in Bob’s brain). Unfortunately you will probably have to figure out what is valuable. Or best guess it. And establish processes & a culture for it to seamlessly get written down.
Tribal knowledge, knowledge silos, the person who was the company brain on topic X had to leave because of an armadillo emergency, Confluence/Docs/pdfs growing too large of a trash heap to find good copper in - these are age old problems companies have worked on. It’s just worse now.
An agent can’t read what was never written down, and neither can the new hire you just employed.
- Efficient. Your docs musn’t discriminate, and be consumable to all stakeholders. Humans and agents alike.
Once it exists, it has to be consumable. To all stakeholders that matter.
If your info sits on paper or in pictures or in not-transcribed phone calls, I’ll hold your hand as I say this : you need to digitalize first my dude. Get that chat online, take your notes on a computer or automate uploading them. Your client prospect doesn’t exist for the automation system if it’s not in Salesforce.
Then think about cost. Human time consuming the info & agent tokens. If the doc is only for humans, keep in mind human cost (time & attention spans) & if it’s for agents (or both), consider tokens.
A wiki written for a human skimming with their biological optical sensors (eyes), pretty but non-direct layout, full of screenshots, or even worse… a process description living in a slide deck. 27 000 PDF files. Agents will either skip these, misunderstand them, or best case scenario - bleed you dry of tokens as you try to process them. Agents and systems will continue to get more efficient and capable every week, but this is the status today.
- Aaand then you have to maintain all of this indefinitely. With culture and processes.
So: Digitalized -> token-efficient format -> organized -> searchable-findable -> owned and maintained
Usually you have to draw a map too. And close your eyes and throw a way a bunch of junk data. And then restore some of that junk data. And build observability around what documentation/context is consumed most and least by both humans and agents.
But then you’re good to go.
Existing, efficient, maintainable. Skip any one of the three and the docs function as a decoration/chore/legal backup, not infrastructure.
Workflows have to be mapped — and agreed on
Ok, I didn’t get to this second part. See you next time!
Neither of these is about the model
This is the same shape as the liability gap I wrote about earlier — the scaffolding to automate a given piece of work usually already exists on the model side. What’s missing is company-side infrastructure: documentation good enough to hand to something that can’t ask a follow-up question in the hallway, and workflows clear enough that there’s actually one version to automate. Solve those two and the “we tried AI and it didn’t really help” story mostly goes away on its own.
I’ll expand the workflow mapping section as well, pinky promise.


