Pablo Arranz retweeted
HOY NO VENGO A OFRECER MIS SERVICIOS. HOY VENGO A PEDIR UNA OPORTUNIDAD.
Tengo 49 años, una esposa y dos hijas que dependen de mí. Desde abril de 2026 estoy desempleado y necesito trabajar.
No es fácil escribir esto, pero hoy me toca pedir una mano. 🧵
Pablo Arranz retweeted
Day 17 on RaceDay.gg: Crystal Ring.
Mrax took it with 29.871, just 0.24 s ahead of Solrac. Pablo Arranz third. 31 drivers on the track.
Top 10 recap: youtu.be/oMMfLKV-Rhs
#raceday #racinggame #gaming
Pablo Arranz retweeted
Live races are now on RaceDay.gg 🏁
Open a room, share the link and race up to 8 friends on the same grid. Real contact, live positions, 1 to 5 laps.
Free, in your browser: raceday.gg
Pablo Arranz retweeted
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
Pablo Arranz retweeted
I've been thinking a lot about the second wave of agents.
Few months ago the question was "can it do the job?"
Now it’s "which version does it best?"
I just wrote a new Engineering Blog post walking through how we actually run A/B experiments on production voice agents and why the loop compounds.
We also dropped a short video on the Agentic Experimentation feature itself.
full write-up
happyrobot.ai/blog/how-ab-te…
Good AI teams ship changes. Great AI teams run experiments.
Every change to how your AI operates has an unknown impact on production. With A/B testing experiments, you can test different versions of your agent against real production traffic simultaneously, before deciding to roll out broadly.
Test a change in prompt details, model choice, or workflow structure - wording, tone, cost profile, logic, steps, or entirely different versions. Measure effect on any custom business metric - customer satisfaction score, recovery rate, escalation rate, conversion, and more, while tracking agent behavior in parallel so you know the variant that wins on performance also behaves exactly as intended.
@gonzalo_io from our engineering team dives deep into how the experimentation capability helps make data driven improvements to your AI workforce: happyrobot.ai/blog/how-ab-te…





