A good agent eval is worth much more than you think.
You can’t improve what you can’t measure, and “it felt better” is not a metric. Write the test harness first. Then iterate for free.
https://t.co/QCsXLOVFQX
model-compatible ≠ model-optimized
Agent harnesses are more model-shaped than model-agnostic.
Anthropic’s Claude Code example makes it clear: Sonnet 4.5 needed context resets for “context anxiety.” Opus 4.5 behaved differently, so those resets could go.
Change the model, and parts of the harness change too.
This makes true model-agnosticism a harder engineering challenge than most frameworks admit.
Automated bug fixing with agents is not one task.
It is two separate runs:
1. Reproduce the bug with a failing test
2. Fix the bug (pass the test)
Better traceability, evaluation leading to reliability.
https://t.co/OvRk2YfqFw
The future of hiring probably looks less like:
“Upload your CV.”
And more like:
“Connect your professional history, and let an agent evaluate fit against the role, team, and company context.”
https://t.co/raguxNBE8r
#HRTech
A positive vibe check is noisy.
A negative vibe check is noisy too.
🤦♂️
One good answer doesn’t mean your agent works.
One bad answer doesn’t mean it’s broken.
🤦♂️
At production scale, you need a structured way to tell the difference.
Evals is the answer
https://t.co/QCsXLOVFQX
@FleihanAline@Stregisdubai I’m so sorry you had to go through this. I hope you’re healing well. There should be a proper intervention from the government to punish them and make sure this doesn’t happen again!! It’s very shameful the way they handled it!!
1/ #Anghami First Arab Tech Company listing on NASDAQ NY
Thread time>
✌🏻How did we go from rejections to stock listing!
It started with a bunch of emails we sent hoping to get an investor engaged. Most remain unanswered, few told us we'll fail. Thankfully one believed.
Excited to announce that WonderEd has been invited to @StartupGrind's Annual Startup Summit, an intensive member-only, half-day bootcamp held the day before #SGGlobal2021 on Feb 23-25!
Thanks @giomate & team!
More info on the Conference here -https://t.co/Of5XB8uGUb
Woohoo! We’ve been accepted into the Startup Grind Membership. Looking for a community that helps startups level up? Check out Startup Grind Membership and join us -
https://t.co/xQRVZEbAVu #StartupGrind#startupgrindmembership#startup
Apple is donating to relief organizations that are helping with immediate needs and long-term support in Beirut. We grieve with the people of Lebanon, our employees and all those affected by the tragedy.
ATTENTION: We are calling on the world, our friends, our fellow Lebanese in the diaspora to DONATE to help us provide disaster relief. We will update the page to provide exact details on how your money will be spent. Donate here: https://t.co/eER5BZs8jG #بيروت
Bridger, 6 years old, saved his little sister from an attacking dog. He knew he would get hurt, but he did it anyway. He’s a hero.
So, we made this happen. One of the most fulfilling things, ever, huge thanks to Chris Evans.
Spread love. ❤️