Quick update from us at 2027.dev, the Agent Experience (AX) platform that helps you ship agent-ready products. July went to a boring-sounding problem with a big payoff: making an eval run in the environment your users actually start from, and telling you plainly when it doesn't.
Prompts that match your real setup
Prompts now carry their own environment:
- A setup command that runs before the agent starts, and shows in the trace as a pre-step with its output
- Environment variables that are write-only: you set them, nothing reads them back
- A sandbox scaffold you pick, so the agent starts where your users start
- A runner pin, if a prompt's scheduled evals should always use one model and harness
Runs that tell you what happened
When a run fails it now says why: the real failure reason, on the run page and in the runs list, instead of a bare failed badge. The status badge carries the eval outcome as well, and traces now separate the errors an agent recovered from on its own from the ones that counted against it.
You can also opt into a browser notification when an in-flight run finishes, so you don't have to sit on the tab.
Landscape: competitors that share your setup
The Competitors tab is now Landscape, but the real change is underneath. A competitor runs as a sub-prompt of yours now, inheriting the parent prompt's env vars and setup command. Before, their run and yours could differ in setup and the comparison quietly stopped being apples to apples. Now it can't.
Slack, now self-serve
Digest cadence moved into your own settings: daily, weekly, off, or auto, which follows however often your prompts run. And the bot installs through a normal Add to Slack flow, so getting it into your workspace no longer needs us in the loop.
Also shipped in July
- Opus 5, Sonnet 5, GPT-5.6, Grok 4.5 and Grok Build added to the eval runner
- Org memory: we synthesise what we've learned about your product across every run, the judge reads it, and you can preview the note in settings
- Download PDF on report pages
Coming soon
Agent Paths: the route agents take through your product, aggregated across every run. Plus a URLs view of every page they fetched to get there.
Thoughts?
We're always here if you have any thoughts on what we can do better. Reach out at mika@2027.dev or DM me on X anytime: @heymikasagi.