If It Isn't Written Down, It Didn't Happen

More companies should steal a simple rule from regulated industries.

If it isn't written down, assume it didn't happen.

That sounds bureaucratic until you've watched an organization try to reconstruct an important decision from six Slack threads, three meetings, two emails, and five people who remember the conversation differently.

I've spent decades on FDA regulated device work, so FDA's documentation instincts are something I've benefitted from in all my startups and fractional CTO work. Shipping higher quality products is not harder with these “write-first, talk-second” rules, it's easier.

FDA folks have been saying some version of this for years. In an FDA training program on corrective and preventive actions, the agency puts it plainly: if it wasn't documented, it didn't occur. Another FDA training program makes the reasoning even clearer. When somebody asks, “Did it happen?”, documentation is the objective evidence that it did. No documentation, no objective evidence. In regulated work, that's the floor, not the slogan.

That's a compliance principle. It's also a good management pattern.

The Meeting Is Not the Record

Most companies operate as if communication itself creates organizational knowledge. It doesn't.

A meeting happened. People talked. Somebody took notes. Someone else remembers the conversation differently. A Slack thread sank under 200 newer messages. Three weeks later somebody says, “I thought we agreed...”

That is communication. It is not governance.

Document First, Then Talk

My preferred pattern is simple:

Document → Discuss → Update → Govern from the document.

This changes what a meeting is for. Once the document exists, the meeting is no longer a place where everyone manufactures another version of the story. The meeting exists to improve the document.

• Someone says a fact is wrong? Fix the document.
• Someone has new evidence? Add it.
• Someone disagrees with the root cause? Record the disagreement and the evidence.
• Someone thinks the fix won't work? Update the fix, or write down why the existing one still stands.
• Someone raises the same complaint two weeks later? Open the document.

This is especially useful when emotions are running high. People are remarkably good at remembering the parts of a conversation that support their position. Documents are less talented at that.

None of this is exotic. Amazon has run on it for decades. In 2004, Bezos sent an email banning PowerPoint from executive meetings; ideas had to show up as narratively structured memos instead. A later shareholder letter describes how those meetings open: everyone sits and silently reads the six-page memo, study-hall style, before a word is spoken. The document comes first, and the discussion serves it. He has called the change the smartest thing they ever did. Steven Sinofsky made the same argument from the Microsoft side in “Writing is thinking”: the act of writing the thing down is where the sloppy thinking gets caught, before the meeting, not during it.

Don't Create Parallel Realities

One of the worst organizational habits is letting an important issue grow several competing histories.

Engineering has its version. Operations has another. The CEO heard something else. Someone summarized the story on Slack. Someone else gave the board the three-minute version.

Now management isn't solving the original problem anymore. Management is trying to figure out which story is true.

The easiest way to prevent this is to declare one artifact authoritative. After that, conversations can challenge the artifact, but they shouldn't compete with it. If somebody says, “That's not what happened,” the answer should be: “Good. Show us what's wrong and let's fix the document.”

This Is Why FDA Cares So Much About Contemporaneous Records

There's another useful lesson buried in FDA data-integrity practice. FDA uses the acronym ALCOA for records: attributable, legible, contemporaneously recorded, original or a true copy, and accurate.

“Contemporaneously recorded” matters a lot. Writing the story six weeks later is not the same thing as writing it while people still know what happened. Memory decays. Incentives change. Outcomes become known. Hindsight starts rewriting the story.

The closer the documentation is to the event or the decision, the more useful it is.

FDA warning letters show why the distinction matters. In one case, a company told FDA it had followed its corrective-action process “yet without the paperwork.” FDA rejected the response, because the required investigation and corrective actions had never been documented. I've sat through enough audits to know that lesson holds whether the FDA ever comes near your company or not.

Writing Is Not Bureaucracy if It Replaces Bureaucracy

Executives sometimes hear “document everything” and picture people spending half their lives filling out forms. That would be stupid.

The goal isn't more documentation. The goal is fewer repeated conversations. A good document eliminates work.

• Instead of explaining the same problem to five people, send the document.
• Instead of holding another status meeting, update the document.
• Instead of asking everyone what happened three months later, read the record.
• Instead of reopening an argument, ask whether there's new evidence. If there is, update the record. If there isn't, move on.

That is less bureaucracy, not more.

AI Makes the Old Excuse Obsolete

Historically, documentation was expensive. Someone had to sit down after the meeting, organize notes, write the report, clean it up, circulate it, collect comments, and revise it.

Modern AI removes most of that cost. Feed it the meeting transcript, the Slack thread, the emails, the logs, the supporting evidence. Ask it to draft the artifact. Have the responsible human correct it. Then make that document the record.

The human judgment still matters. The typing doesn't.

That changes the economics of organizational discipline. There is almost no good excuse left for an important meeting, incident, engineering concern, customer escalation, or management decision to exist only as conversation. And yes, I use AI for exactly this kind of drafting, the same way I use it for my own writing: the machine drafts, the human owns the result.

If you want a companion pattern for what happens to the conversations themselves, I wrote about instrumenting them in Stop Telling People to Communicate Better. Instrument It.

AI Can't Work from Human Memory Either

The cost argument is only half of it. The other half matters more.

AI has no access to your organization's memory. It can't read a hallway conversation, a Slack thread that sank, or five people who remember the meeting differently. It can only work from what's written down.

A decision that lives only as conversation is invisible to AI. A decision that lives in a document is something AI can read, summarize, search, reason about, and turn into the next artifact. Writing is how the machine gets the context it needs to do anything useful for you.

Before AI, an undocumented decision just became hard to reconstruct. Now it's also outside your AI tools' reach entirely. The rule matters more, not less.

So the old line gets a second half: if it isn't written down, it didn't happen. And if it isn't written down, AI can't help you with it either.

Small Companies Need This More Than Large Companies

Founders sometimes tell me, “We're too small for all this process.” I think they have it backwards.

Small organizations run on oral history. Everybody knows everybody. Important context lives in people's heads. Decisions get made fast. Slack feels like documentation because everyone remembers the conversation.

Then the company grows. A few people leave. Customers get larger. Problems get more consequential. Suddenly nobody can remember why something was done six months ago.

You don't need enterprise bureaucracy at 30 employees. You do need organizational memory. Those are different things.

Use Documents as Organizational Memory

I'd apply this pattern anywhere the cost of forgetting, misunderstanding, or retelling something is real.

• An engineering concern gets an Engineering Concern document.
• A serious incident gets an RCA and a remediation record.
• A security finding gets a finding and a remediation plan.
• An important product decision gets a decision record.
• A quality problem gets the appropriate quality-system record.
• A major management decision gets a short decision memo.

Then apply one rule consistently: once the authoritative record exists, future conversations improve the record rather than compete with it.

Writing first, talking second. And once the writing exists, the talking should improve the writing.

That isn't just an FDA habit. It's a pretty good way to run a company.