Technical Co-founder at Ansight.
Sydney, Australia.
Joined September 2015
- Tweets2.8K
- Following76
- Followers2.9K
- Likes4.6K
Did you actually review your last AI-generated PR, or just ask the same agent for a summary?
Code generation has now far outpaced human attention.
At Ansight, we have an alternative philosophy: Evidence Engineering.
ansight.ai/articles/evidence…
Checking the same app journeys by hand before every release? 👀
UI tests still on the backlog because they’re a PITA to build? 🤦♂️
Use Ansight to turn easily create UI tests, written in plain language, from an existing app session replay.
ansight.ai/articles/from-man…
Very pleased to announce that the Ansight CLI v1 is out 🎉
Connect a coding agent to a running mobile app to capture, replay, and test with a session you can review as proof of work.
Read the announcement here: ansight.ai/articles/introduc…
Matthew Robbins retweeted
"Done" is a claim. A session recording is evidence.
Attach the Ansight session to the PR: what the app rendered, requested, logged and stored at every step of the run. Reviewers see what the app did instead of trusting what the agent said.
ansight.ai/
Matthew Robbins retweeted
The same mobile test, 4.2× faster with Ansight.
Same four tasks, same app. A computer-use agent (Sol) took 120.6 seconds, working from pixels tap by tap. On Ansight the agent ran the stable steps as code and read a UI tree: 28.7 seconds. Results vary.
ansight.ai/#efficiency
Matthew Robbins retweeted
The agent says it verified the flow. But did you?
Ansight did. It files a screenshot and UI tree before and after each action, hashed. A fuzzy match can be read but never tapped. OCR is never proof of absence.
Claims come with receipts.
ansight.ai/
Rebuilding a @maestro__dev test tap by tap when the journey’s already recorded is a little bit silly. 🙃
We built Ansight to extract a draft from the replay. Review the selectors, add an assertion, get on with your day. 😉
ansight.ai/articles/your-ui-…
Matthew Robbins retweeted
A crash report shows where it died. Not what happened.
Reproduce the crash with Ansight and the run-up is recorded: screens, taps, logs, requests before it went down. The agent reads what led there, fixes that, and reproduces again to prove it.
ansight.ai/
Matthew Robbins retweeted
Passing tests are not a working feature.
The email test passes. The mobile app rejects a valid address.
Ansight lets the agent read the form's runtime state, fix the flow and verify it works before handing it over.
ansight.ai/
Matthew Robbins retweeted
Agentic testing shouldn't eat your coding token budget.
Test every change without burning the tokens your engineers code with. Stable steps run as code at zero tokens and the agent reads only what it needs. 94.7% fewer tokens in our benchmark.
ansight.ai/#efficiency
Matthew Robbins retweeted
Connect once. Every mobile app run, captured.
Once connected, Ansight captures dev sessions and UI tests automatically.
Simulators, emulators, real devices. Runtime evidence, metadata and insights, ready to review.
ansight.ai/
Matthew Robbins retweeted
The feature works. Memory is up 40 MB and nobody looked.
Ansight grades every finished session against your thresholds: retained memory and frame rate, per journey. A regression needs two confirming runs against ten comparable ones before it flags.
ansight.ai/
Matthew Robbins retweeted
The agent opened the PR. You're still the one who has to verify it.
With Ansight the run comes with the PR: the build on a device in CI, the journeys exercised, what held and what didn't, with the screen and data behind each step.
Proven, or flagged.
ansight.ai/
Matthew Robbins retweeted
Your coding agent is spending its token budget driving your mobile app, one action at a time.
Every tap is a turn. On Ansight it calls a small automation for the steps already solved and reads only what it needs for the rest. Far fewer turns per test.
ansight.ai/
Matthew Robbins retweeted
A label changed. Time to rebuild your test suite 🫠
Ansight is mixed-approach testing: your agent runs the test and calls short automated Tasks for the solved steps. Only what changed needs judgement, and the outcome is verified from the app's state.
ansight.ai/
We've had HUGE breakthroughs in @Ansight_ai, our agentic dev tool apps. 🔥
Against computer use:
20x less token use.
4x faster test execution.
~98% cheaper per test.
Big things coming; use those precious tokens you're spending testing your app to actually build it instead. 😉
Does this sound familiar?
You find a bug in your mobile app, lose the exact repro, paste screenshots/logs into Codex/Claude, and wait while it guesses from source code.
Ansight fixes that with full-fidelity replay, runtime inspection, and app control.
ansight.ai/blog/introducing-…
After a few years, I have managed to return to my roots of building developer tools.
@Martinsmora3 and I are super excited about @Ansight_ai and the productivity gains we have already experienced.
Matthew Robbins retweeted
Ansight’s Beta is finally public now!
@matthewrdev and @Martinsmora3 built Ansight because Coding Agents are blind on the other side of the Compiler.
We solved that problem by turning Mobile and Desktop Apps into a structured tool surface for Agents.
ansight.ai/blog/introducing-…