In college, my QA professor asked us:
"Would an engineer prove a building won't fall by gently leaning on it?"
10+ years later, I still think about it
Testing isn't proving your code works
It's trying hard to break it
Inputs you'd never type
Load you'd never expect
Users who don't think like you
Your tests have to go beyond the way you think
@NamAniket I've got access to Duet few days ago, it is still more like a prototype, it lacks good UX for instance, but it has access to the good models like the GOAT Opus 5.5
Grok Bot was the best harness for software engineering I've used
Then the subscription changed, no more Opus 5.5 model included, so I stopped using it for SWE
Now I'm tying to build that "harness" in Claude Code
Has anyone gotten Claude Code to feel even close to Grok Bot harness?
Dozens of pull requests being merged automatically
This wasn't easy a year ago for a small team
Now it's viable thanks to several guardrails and lots of automated deterministic checks
Microsoft 365 has been down since yesterday. Outlook, Teams, SharePoint.
If a giant can stall, so can we
- A disaster recovery plan team never tested is a wish
- Simple architectures recover faster
- Design for failure: redundancy, failover, multi-cloud
- Defensive programming (we easily underestimate the power of this)
Which one does your team skip?