Hello from Morse: what the product posts will cover
We're opening Morse in a soft launch this October. Here is the question the product is built around, what a performance brief is for, and what you'll read in this lane.
By The Morse team
Most teams we talk to can tell you what their agents cost last month. Far fewer can tell you whether those agents got better — and almost nobody can say both in the same sentence.
That gap is why Morse exists. This October we're opening it in a soft launch, to a small group of teams running agents in production, and starting this blog alongside it.
The question behind the product
Morse is AI-agent observability and evals, built around one question: which agents are getting better, and which are just getting expensive?
The answer arrives as a performance brief. It runs one loop:
- Measure what each agent costs and how often it actually succeeds in production.
- Prioritize the few agents and workflows that need a person to look at them.
- Verify that the change you made moved the number you cared about.
A brief starts from a short list of facts per agent:
| Signal | What it tells you |
|---|---|
| Spend | What the agent costs per run and per successful outcome |
| Production success | Whether labeled outcomes are trending up or down |
| Sampled quality | Separate, sampled quality checks — never fused into one score |
Every line in a brief links back to the traces behind it. Performance brief first; traces on demand.
What the product lane covers
Product posts are about what Morse does and why it does it that way:
- What shipped, described plainly, including what it does not do yet.
- What we measured on our own agents — Morse runs on Morse.
- What we changed our minds about, and the evidence that changed it.
The first two are already up: what a performance brief is (and what it is not) and why we lead with the brief instead of a wall of traces.
The other two lanes
- Engineering — how we build Morse: typed spans, ingestion, trace-anchored correlation, and the trade-offs behind them.
- Industry — where agent observability and evals are heading, and the practices we think hold up in production.
Getting the facts is mostly plumbing
Instrumenting an agent is two lines:
pip install morse-aiimport morse_ai
# Reads MORSE_API_KEY from your environment
morse_ai.init()The interesting part is what you do once you have the facts — and that's what most of these posts will be about.
Follow along
Subscribe via RSS, or join the waitlist to get a performance brief on your own agents.