Stop writing reports — put it in the knowledge base. Bureaucracy in the age of AI

2026-06-30 · Ascendy Engineering

TL;DR

A big company’s “reporting for the sake of reporting” — the time and people spent polishing the report — should go. Some even use AI to do it better, which is using AI in the wrong place: a pure waste of tokens.
A report is push — the sender packages it to the receiver’s taste and pushes it. That’s a huge chunk of bureaucratic time.
The alternative is pull. Humans drop raw data, with its context, into a knowledge base, and an agent processes, organizes, and makes it queryable. Whoever needs it takes it in their own format. The “piled-up = garbage” trap is solved not by a person who curates but by an agent that does.
So big companies, inevitably inefficient, split into small units. But they don’t disappear — they keep their strong domains, while small AI-leveraged firms take the weak ones.

About this piece. The view of someone who works inside a large organization while also building a product solo — crossing both worlds. Company name and affiliation are generalized. Predictions about the future of organizations, and strong claims like “left behind,” are bets, not facts (N=1). Same vein as spend fewer tokens, spend more and Build in studying.

The irony of using AI to write a better report

I work inside a large organization while building a product solo. Crossing both worlds daily, one contrast is stark: solo work has no reporting. A big company has a lot of it.

What grates most is reporting for the sake of reporting — time and people spent not on the content but on making the report better. Re-shaping the same information to a superior’s taste, prettifying slides, tuning the tone. And lately — people use AI to do that better.

That’s the irony. Using AI to dress up reports is using AI in the wrong place — pure waste. Burning tokens to do labor that should disappear, faster. The real question isn’t “how do I write a better report?” It’s “is this report even necessary?”

A report is push — and that’s the root of the inefficiency

Step back and a report is, at heart, information sharing — making what I know known to others. But the report form does it as push: the sender imagines the receiver, packages it to their taste, and pushes it out.

That’s where the cost is. The bulk of writing a report isn’t producing the information but packaging it for the receiver. You rewrite the same content for the exec, the lead, the neighboring team. And most reports are read once and discarded. The packaging made for sharing becomes garbage the moment the sharing ends.

A large share of bureaucratic time is this packaging labor. And AI should be used to eliminate it — not to accelerate it.

The alternative is pull — but an agent must be the processing layer

So what do you do? Flip push to pull. Instead of the sender packaging and pushing, let whoever needs it take it, when they need it, in the format they need.

Here’s how. Instead of writing a report, humans drop raw data, with its context, into a knowledge base — what happened, what numbers came out, what was decided and why, unpackaged. And the reader takes it in the format they need: the exec a summary, the engineer the detail, the neighboring team a version fit to their context.

The obvious objection: “isn’t a pile of raw data just garbage?” Yes. Unorganized information isn’t searchable and isn’t trustworthy. Traditionally that’s why you needed a person to curate — a doc manager, a wiki gardener.

This is the crux. Now an agent is that processing and curation layer. Humans only put data in; the agent indexes it, links it, and assembles it into the format a question asks for. The old pull trap — “piled up means garbage” — is solved not by a curating person but by a curating agent. This is what using AI in the right place looks like: not accelerating the packaging, but removing the need to package.

So bureaucracy gets restructured

If reporting disappears, what’s left? Much of the bureaucratic layer built on reporting, coordination, and approvals was about moving information between people and gating it. When a big chunk of that moves to the knowledge base and the agent layer, the layer itself thins.

My position is clear. A big-company structure is inevitably inefficient and must be split into very small units of work. Much of a large org’s inefficiency is “coordination cost from having many people”; when AI absorbs a large part of that coordination, the reason to stay big shrinks.

The uncomfortable part — people who can’t use AI

Here’s the uncomfortable bit. Sadly, I think people who can’t make use of AI become subject to adjustment. AI is like the computer. The same thing that makes it nearly impossible today to hold an office job without being able to use a computer — I think that happens with AI.

This is an observation about structure, not a condemnation of people. And it’s a bet from one end — a solo operator’s — not a proven proposition (hence N=1). Like the computer, AI is ultimately a tool you can learn, and the way to prevent being left behind isn’t to blame people but to build the chance to learn into the structure. What’s fading is the option of not learning — that’s the direction I see.

So do big companies disappear? No

Here’s the balance. “With AI, everything becomes solo” is naive. There are clearly things I can’t do well solo — branding, marketing, design, where there’s a real gap between me doing it clumsily with AI’s help and an expert in that field using AI. Expert + AI is almost always better.

So the future I see isn’t “the death of big companies.” Big companies keep what they’re strong at — capital, scale, a pool of experts. But the weak parts, the domains where inefficiency piled up, get taken one by one by small firms that maxed out their leverage with AI. The big org thins to its strengths, and high-leverage small cells fill the space it vacates.

Takeaways

Don’t use AI to write better reports. That accelerates labor that should disappear — a waste of tokens. Use AI to eliminate the report form itself.
Flip push (reports) to pull (a knowledge base). But the “piled-up = garbage” trap only resolves if a curating agent is the layer. Humans put data in; the agent processes it.
Organizations shrink to the unit of AI leverage. Big companies don’t vanish but thin to their strengths; small high-leverage firms take the weak parts. “Everything becomes solo” is naive — because expert + AI is almost always better.

Authorship & citation: Written by Ascendy Engineering; quotable with attribution. Found something wrong? Let us know via a GitHub issue.

Tags: ai, future-of-work, organization, bureaucracy, knowledge-management, opinion