Short answer: a consumer products team asked us whether AI could write the detailed documents they prepare for every new product. After sitting with the people who do that work, we built them something simpler: a small custom tool shaped around how they already work. The re-typing step is gone entirely: the details come straight from the files the team already works in. There's no AI subscription to pay and no server to maintain.
The ask: "Can AI do this for us?"
Every new product meant a long, detailed document that other companies would rely on. The information lived in a few places: the design files, a large spreadsheet template, and a lot of back-and-forth messages as the document passed between teams. It was slow, the template broke easily, and nobody was ever quite sure which version was the latest.
They'd seen what AI can do with text, so the question was natural: could AI just write these for them?
What we saw when we sat with the team
We spent time with the people doing the work, watching how a document actually got made. The hard part wasn't writing. Almost everything that went into the document already existed in their files. The hours went into:
- re-typing the same details from one file into another;
- fixing the template when an edit broke its layout;
- chasing the latest version as the work moved from one team to the next.
That's a data-flow problem, not a writing problem. An AI tool would have added a monthly bill and a new source of mistakes, and the re-typing would still be there.
What we built instead
A focused desktop tool, built around their existing workflow rather than a new one:
- It reads what already exists. Details come straight from the team's working files, so nothing gets typed twice.
- One place for each document. Everyone works on the same version instead of trading copies.
- Clean hand-offs. When one team finishes, the document moves to the next with its status attached.
- Their formats, unchanged. It produces the exact files the people downstream already expect, so nobody outside the team had to change anything.
- Nothing to babysit. It runs on the team's own computers: no server, no per-seat fees, no AI subscription.
It wasn't a big platform or a complex AI system. It was a straightforward build, and that's why it worked.
Where AI did help
We still used AI where language was genuinely involved: a small helper that turns rough spoken notes into consistent written comments, with a person approving every output before it goes anywhere. The point isn't "no AI." It's AI where it pays for itself, plain software everywhere else.
Is your problem a language problem or a data-flow problem?
| Signal | Points to a simple custom tool | Points to AI |
|---|---|---|
| Where the information comes from | It already exists in files or systems | It has to be written, summarized, or interpreted |
| What goes wrong today | Re-typing, version mix-ups, broken templates | Drafting takes too long or needs expertise |
| What a correct output looks like | Exactly matches the source | Several good answers are acceptable |
| Cost of an error | High: wrong details flow downstream | Lower: a person reviews and edits |
Simple tool vs. AI subscription
| AI tool (what was asked for) | Simple custom tool (what we built) | |
|---|---|---|
| Ongoing cost | Monthly subscription, indefinitely | Normal upkeep only; runs on their own computers |
| Fit to the workflow | Generic; the team adapts to it | Built around their files, formats, and hand-offs |
| Accuracy | Every output needs checking | Reads details straight from the source |
| Ownership | Rented | Owned by the business |
Before you buy an AI tool, ask
- Where do the hours actually go? Watch the work for a day. Count typing versus thinking.
- Does the information already exist? If yes, moving it reliably beats generating it.
- What will this cost in year three? Subscriptions are forever; a simple tool can run for years.
- Will your partner tell you no? The best engineers sometimes talk you out of what you asked for. Our build vs. buy assessment walks through the same judgment.
Work with First Mate
First Mate is a San Francisco-headquartered engineering partner for startups: Silicon Valley-trained leadership, senior product-minded engineers in Manila, and a two-week paid trial to prove the fit. See how First Mate builds custom software or start your two-week trial.
Building something with AI?
First Mate is an AI engineer agency. Our product-minded AI engineers scope the problem before they build it. You can start with a rough cost and timeline, or talk to us directly.
