attentionfactory

We build our own software.Then we tell you what it cost.

The Lab is where we ship products, not decks. Some run in production. Some are experiments we keep in public because the failures are useful.

TranscriptX.Live, and paying its own way.

Live

Turns recorded conversations into searchable, quotable text.

Built for teams who record everything and read none of it: interviews, client calls, research sessions and workshops become text you can search on Monday instead of audio nobody opens again.

Four more, at four stages.We label them honestly.

In beta

Signal Desk

Reads the week in AI and drafts the briefing.

It is how our Intelligence team keeps up with releases, papers and pricing changes without spending two days a week reading.

In beta

Runbook

Turns a messy process document into a working automation.

We use it on the first week of a build to get from someone's Google Doc to something that actually runs.

Experiment

Groundskeeper

Checks agent output against the source documents.

An answer to the one question every client asks us: how do you know the thing did not make that up.

Experiment

Halfpipe

Scores prompts before they reach production.

A test harness for the prompts inside our own client work, so a rewrite gets checked instead of eyeballed.

No download counts. No adoption charts. We have not measured those, so we are not printing them.

Most AI firms have never maintained an AI product.It shows in the advice.

Reason 01

We hit the walls before you do.

Evaluation, permissions, cost per run, the moment a model changes underneath you. We have paid for those lessons on our own roadmap, not on your budget.

Reason 02

Advice from people who ship is different advice.

A firm that has never maintained its own software will tell you what is possible. A firm that maintains five things will tell you what it costs to keep them running in month nine.

Reason 03

The tools go into the client work.

Runbook and Halfpipe were built for our own delivery. When they earn their place internally, they show up in your build too, already tested.

Want one of these pointed at your problem?

Tell us what your team keeps doing by hand. If something on this shelf fits, we will say so. If it does not, we will say that instead.

  • Every product starts as a tool we needed for a client build.
  • Experiments stay labelled experiments until they earn a promotion.
  • Anything we retire, we say we retired.