Alex Haefner

@lg Interesting setup! I’d rather the harness do what you describe. Plan mode in codex is straight up broken, mostly ditched using it. Some folks suggested I asked questions to codex multiple times to build the context then ask it what to do without using plans. Seems to work better

@alexhaefner the slowness isn't great, but these days i'm getting all implement plans reviewed by fable, codex, kimi, muse and grok. something unique always comes up, misses very little. not too hard to do, main agent calls commandline of others. i do like grok's speed though...

As a former software engineer, I decided to try all of the coding agents this year to see how they compare and where hype meets reality. So far in my testing cursor with composer 2/2.5 beat out codex with 5.5/5.6 by a wide margin. Codex repeatedly suffered from problems…

It took a lot of work to get it to work and this was on one of the most expensive plans. I’m not sure the economics of it make sense right now until it gets better, faster, and cheaper. Onto Claude code next!

(6) lastly, it can do lazy implementations that don’t follow good software design practices. It build a backend/frontend system for a basic queue that fully saturated the db with a single client. It kept everything in memory for machine learning tasks until memory was exhausted…

By posting on X you’re helping launch rockets to mars and building out AI. Don’t let anyone ever tell you otherwise

@Scott_Wiener Hopefully you’re referring to MPU misleading their readers. Liberty, CPUC, and the folks trying to bloc new transmission need to figure out their development plans or find a compromise with NV energy because this was agreed to a long time ago and the NIMBYism has stopped it

This is shameful & should be illegal.

@bgurley Audits are extremely repetitive and lots of ops. Why wouldn’t you assume a startup comes into this space? AI to collect all the data, interview users, get everything into the right form then a human to review and sign off. pass on a portion of the improved margin as savings

We all know with certainty that AI will drastically reduce the costs of corporate auditing. This cost-reduction already widespread in legal. One problem - the big 4 are holding the line, telling clients despite AI, we are not lowering rates. Curious how this lands. The "task" is way cheaper. Will one of the "next three" start disrupting with price? Will the PCAOB circle ranks in an ugly act of regulatory capture? I'm watching with interest. Any predictions?

X seems to be serving ads in your feed without attribution. It’s extreme enshittificafion. Here’s an example of an ad served to me on X for someone I don’t follow in the “following” tab with no marked ad tag anywhere

Anyone else notice X is putting ads in your following feed without any clear notification / moniker? Not people you’re following, no “Ad” pill on them, just for things people / companies are promoting, not clearly called out as ads. This has been banned by the FTC, why now?

“People” and “friends” were just agents with computers the whole time

"But here is what we found when we tested: We took the specific vulnerabilities Anthropic showcases in their announcement, isolated the relevant code, and ran them through small, cheap, open-weights models. Those models recovered much of the same analysis. Eight out of eight models detected Mythos's flagship FreeBSD exploit, including one with only 3.6 billion active parameters costing $0.11 per million tokens. A 5.1B-active open model recovered the core chain of the 27-year-old OpenBSD bug." aisle.com/blog/ai-cybers…

Ai agent coding is gambling for software engineers. Submit a prompt -> wait -> reward or lose. Explains some of the burn out effects and submitting prompts at 2am

@ParthaSundaram I offered ChatGPT I’d pay for the best plan if it could pick the best deal for me on some replacement appliances. It didn’t take the bait (clear objective) and sent me a list of bullet points with emojis and no obvious answer to the core question asked.

cant cancel chatgpt on chatgpt