Our thinking
Insights
What we have learned building AI systems that operators actually run. Reports, articles and perspectives, written by the people who did the work.
All insights
Perspective · Agentic systems
Agents that do office work: what actually holds up in production
After building agent systems for operators in several industries, the pattern that survives is not the clever prompt. It is a deterministic core, a model at the edges, a person one click away, and one file that says who the business is.
Read the perspective
Report · Energy, climate & carbon markets
What a carbon credit actually costs
The listed price is not the price. The brochure is not the integrity. And the trade history, which is public, is the number nobody looks at. A sourcing engine built from public venues shows what a buyer should actually expect to pay for a credible tonne.
Read the report
Article · Logistics
The pricing engine should not be the language model
A dispatcher pasting a load-board posting into a chatbot gets a confident number and no way to know where it came from. The fix is not a better prompt. It is a quote built line by line by code, with the model limited to reading and reviewing.
Read the article
Report · Healthcare & life sciences
Grading a segmentation mask when nobody knows the right answer
Judging a mask against an expert's is arithmetic. The case teams actually face is a model producing thousands of masks with no reference at all. The honest answer is triage: a score, a confidence band that decides the colour, and a queue sorted worst first.
Read the report
Article · Healthcare & life sciences
Screening hundreds of standards without trusting the model
A regulatory team walks a workbook of recognised consensus standards for every submission and fills four columns by hand. An assistant can do the first pass in an afternoon, provided it shows its tools, runs a rule-based second opinion, and gives the workbook back exactly as it found it.
Read the article
Article · Professional services
The email that mentions two jobs
The hardest email to file is the one whose subject names one project while the body discusses another. A filing agent that guesses will be wrong in the way that matters most. One that asks will be right, and will learn.
Read the article
Article · Voice & conversational AI
Voice agents at the front desk: latency, interruption, and knowing when to hand off
A phone agent is judged in the first two seconds. What makes one usable is not the model's vocabulary but the plumbing: no transcoding, interruption that works, a person rung first, and one brain shared with the website so the answers match.
Read the article
Article · Logistics
A certificate forwarded by the carrier never passes
Automating a document workflow means deciding who is allowed to be the source of each document. In carrier onboarding, the one document that proves insurance is the one the carrier must not be allowed to supply.
Read the article
Article · Evaluation & assurance
Measuring a grounded assistant: context recall and faithfulness
"Does it make things up?" is the only question a business owner asks about an assistant built on their own content. It has a numeric answer, and the number should be on the dashboard before the assistant is on the phone.
Read the article
Perspective · Strategy
The AI roadmap for a forty-person company
Most AI advice is written for companies with a data team. For an operator with forty people, the roadmap is shorter and more concrete: find the workflow that eats the most hours, build one system that gives them back, measure it, and only then talk about transformation.
Read the perspectiveNothing matches that. Try a broader word, or ask us directly.
By topic
Talk to the people who wrote these
Every piece describes a system we have built. We can show it running.