Notes on answering people well
Product updates, and longer pieces on support, automation and the craft of a good answer.
What we shipped in July
A calmer onboarding, file understanding across chat and email, and a plugin that puts Tagni on your site in one click.
4 min readCustomer onboarding in 2026: the activation playbook
Retention is won or lost in the first fortnight. The stages that matter, the metrics worth watching, and where support quietly decides the outcome.
The support metrics that actually mean something
Which numbers deserve a place on the wall, which ones quietly mislead you, and how to read them together so the dashboard tells the truth.
Deflection is not resolution
The most flattering number in support automation is also the least honest. Here is how to tell the two apart, and why the difference compounds.
How we keep agents from making things up
Grounding, refusal, and the unglamorous engineering that decides whether an answer can be trusted in front of a paying customer.
Writing documentation your AI can actually use
Most help centres are written for a reader who already knows what they are looking for. Here is what to change so an agent can answer from them.
What we shipped in June
Per agent tone controls, a much faster first crawl, conflict detection in your sources, and a reporting export people had been asking for since March.
Written for people who answer people
We publish what we learn running support at volume: what moved the numbers, what quietly broke, and the engineering behind answers you can actually trust.
- UpdatesWhat shipped, every month
- InsightsMeasurement, retention and the honest numbers
- EngineeringHow the agents are actually built
- GuidesPractical work you can do this week