01
Operational AI agents
Agents that read, search, draft, update tools and escalate with context. Built around a narrow workflow, not a vague promise.
Independent AI studio · Bordeaux / remote
Useful AI, built into the work.
Last Word designs and ships agents, automations, scraping systems and websites that work inside real teams and real tools.
01 / Work
We look at the real workflow before choosing the technology.
02 / Agents
Agents that search, sort, draft, monitor and escalate.
03 / Delivery
Connected to your tools. Tested on real cases.
Point of view
The valuable part of AI is rarely the chat window. It is the workflow around it.
Teams do not need another spectacular prompt. They need fewer repeated checks, fewer copy-pastes, cleaner handoffs, better monitoring, and tools that say when they are unsure.
We take one recurring workflow, understand how it really behaves, then build the smallest reliable system that can absorb it without creating a new mess elsewhere.
02
Automation for messy work
Reporting, follow-ups, handovers, document handling and back-office routines turned into observable systems.
03
Market watch & web scraping
Collection, monitoring and dashboards with traceable sources, explicit gaps and alerts humans can trust.
04
Creative web systems
Editorial websites and landing pages with sharp copy, fast pages and a visual direction that does not look like every AI startup.
A proof you can inspect
We started with our own website.
No client data and no borrowed success story. We ran a fixed set of French and English pages through the same checks at desktop and mobile sizes. The run left timestamped records, a review matrix and hashed files. A separate reviewer still decides whether the work is fit to ship.
Redacted outputs, protocol and limits. No client material.
Problem
Check a bilingual journey without relying on a screenshot or a vague “works for me”.
Inputs
Public URLs on both domains, desktop and mobile viewports, and a test list written before the run.
System
A headless browser records status, final URL, canonicals, language links, console output, failed requests and rendered structure.
Output
One JSON record per page and viewport, a review matrix and a hash manifest that exposes changed evidence.
Limit
This is a point-in-time observation. We did not submit a form or infer conversion performance from the run.
Human review
A separate reviewer checks the findings, separates evidence from assumptions and can stop the release.
How we work
First the sources. Then the system. Then a clear decision on what to build, stop or improve.
01
Map the work
We start with the actual sources, exceptions, decisions and handoffs. The shape of the system comes from the work, not from the model.
02
Build a small proof
A first version must run on real cases, expose its limits and make the next decision easier. No staged demos, no fake data.
03
Make it observable
Logs, review points, failure states and human override are part of the product. If nobody can supervise it, it is not ready.
The last word
Bring the awkward process. We will make it testable.
A vague idea, a painful routine, a monitoring need, a site that needs a point of view: that is enough to start.
Start with one workflow