GPT-6 Astra in Codex is helping test code end to end at Perplexity. Johnny Ho uses it to build test harnesses and mock third-party API responses, so he can check how the pieces work together.

@perplexity_ai @randomjohnnyh How about bringing 4o back into GPT, give people what they actually want #4oForAll

@perplexity_ai @randomjohnnyh it's amazing at its job, but even the people who built it are a little scared of it.

@perplexity_ai @randomjohnnyh When are you guys opening up the $200 subscriptions again?

@perplexity_ai @randomjohnnyh

@perplexity_ai @randomjohnnyh Astra is unusable on $200 plan.I got less than one day of usage on it.

@perplexity_ai @randomjohnnyh Are you kidding me? Fix the fucking model before you start posting trash like this.

@perplexity_ai @randomjohnnyh Perplexity still exists?

@perplexity_ai @randomjohnnyh Mock'lar gerçek API davranışını taklit ederken, uçtan uca testlerde yanlış güven riskini nasıl önlüyorsunuz?

@perplexity_ai @randomjohnnyh Wait perplexity still exist?!

@perplexity_ai @randomjohnnyh Does it mock API errors?

@perplexity_ai @randomjohnnyh End-to-end testing plus mocked third-party responses is a much better Codex demo than code generation alone—the harness is where you find the integration failures users actually feel.

@perplexity_ai @randomjohnnyh any chance of getting more than 1 credit

@perplexity_ai @randomjohnnyh End-to-end testing with AI could dramatically shorten the gap between writing code and shipping reliable products.

@perplexity_ai @randomjohnnyh

@perplexity_ai @randomjohnnyh 端到端测试里把第三方 API mock 住，确实比拿真实接口碰运气靠谱。agent 还得自己跑完测试链路这一点也很关键，能早一点暴露那些“每个函数都对，拼起来却不通”的问题。

@perplexity_ai @randomjohnnyh so astras now generating test mocks and harnesses. that's genuinely useful mocking external apis is tedious boilerplate that ai is actually good at. question is whether the mocks catch the edge cases or if devs still have to debug the same bugs they would've found manually anyway

@perplexity_ai @randomjohnnyh Traction is clearly real on both sides. The economics diverged hard this quarter though — Anthropic did $11.5B at +$559M, OpenAI $6.7B at -$12.3B.

@perplexity_ai @randomjohnnyh Astra mocking the APIs so Perplexity doesn’t have to mock their own integration tests. Progress.

@perplexity_ai @randomjohnnyh The impressive part isn't Astra writing tests — it's Astra knowing what should break. That's the line between "looks like code" and "is actually software."

@perplexity_ai @randomjohnnyh Astra is cooking

@perplexity_ai @randomjohnnyh Nice — mocking third-party APIs for real e2e tests is exactly where this shines.

@perplexity_ai @randomjohnnyh @randomjohnnyh how many w/pm ??

@perplexity_ai @randomjohnnyh mocking third-party APIs is where LLMs actually save time over writing boilerplate manually

@perplexity_ai @randomjohnnyh writing mocks was never the hard part of e2e testing

@perplexity_ai @randomjohnnyh the agent learned mocks. production is now the only environment left that can hurt it