Products
Products and agents
Alongside client engagements we build a small number of productized agents and tools. They are live things you can use, which is a more honest form of capability proof than a case study.
What is here
Superagent. A general-purpose agent built on the Claude Agent SDK. It takes an outcome rather than a prompt, works the task through, and parks anything irreversible for a person to release. In development, and running on our own work.
SEO Operations Hub. Rankings, markets, landing pages, data health and link building on one surface, with an approvals queue in front of anything that touches a live site. This is what Backlink Champion grew into. In development.
AI ROI Scorecard. A free calculator that estimates the annual value of capacity an AI implementation could reclaim, using your own numbers. It shows its formula and states what it deliberately excludes.
BT Labs. Our applied research arm. Frameworks, verified data and field notes on AI implementation, including the results that went against us.
What we mean by an agent
An agent, in our usage, takes an outcome rather than a prompt. It works through multiple steps, uses tools or data sources on its own, and returns a finished deliverable that a person reviews before anything is acted on. That last clause is the important one. Every agent we ship keeps a human approval gate in front of publishing, spending, sending or changing a live system.
We also build custom agents for clients, on open-source frameworks the client owns. That work is described under services, and the frameworks we choose between are defined in the methodology glossary.
How we decide what to build
We build a product when the same piece of work shows up across enough client engagements that automating it is cheaper than repeating it, and when the output can be checked by the person receiving it. If we cannot see how a user would verify what the agent returned, we do not ship it.
Everything we build for a client, they own: code, data and configuration. The productized agents are the exception, and we say so plainly rather than blurring the two.
Custom work
Most of what we build is not on this page. When an off-the-shelf tool cannot do what a client needs, we build one on open-source frameworks the client owns. That work is described under services.
Start the conversation
The first call is a conversation, not a pitch.
Sixty minutes. You tell us what you have and what you are trying to fix. We tell you whether we can help. If we cannot, we will say so.