Dmitri Pisarev

@simonw I do. But I don't read those tests. It's a way for the agent to verify its own work, I treat tests as a manager: you do you in order to verify what you have built

Do you have your coding agents include automated tests for the code that they write?

@FatherofCountry @SpursCulture lol my thoughts exactly x.com/dimaip/status/…

@SpursCulture Thunder -vs- Spurs is the real Finals. Would have been and OKC sweep if OKC was fully healthy. OKC down two of its best players gives, and San Antonio having a historic advantage at the FT line (+17), gives San Antonio a good chance.

Non-Spurs and non-Thunder fans, what Finals matchup do y'all want to see? Knicks vs Spurs Knicks vs Thunder

@SpursCulture Finals are already happening now: it’s Spurs vs Thunder. The actual finals are secondary.

@Swizec @DavidKPiano I guess there are levels to everthing 😀

@dimaip @DavidKPiano They do that also. But you should see the awful jQuery they write ;)

Agents work well with static types, schemas, state machines, etc. Keep deterministic what should be deterministic. Stop trying to make everything non-deterministic.

@Swizec @DavidKPiano I envy you. My agents write weird useEffect heavy code and I have to scold it quite a bit to get state handling right

@DavidKPiano Agents do really well in React codebases because there’s a lot of guarantees. All side effects are declared, everything is idempotent and contained, it’s great.

@DavidKPiano Agents are very close to humans. What worked well for humans, works well for agents

@bridgemindai What's the use for such small models?

MacBook Pro M5 Max is fully set up and running local models. I have never seen speeds this fast. Qwen 3.6 35B and Gemma 4 31B are running blazing fast on 128GB of unified memory. Faster than both of my stacked NVIDIA DGX Sparks sitting right next to it. Initial impressions: Apple silicon in 2026 is no joke for local inference. The M5 Max handles 30B+ parameter models like they're nothing. Full review and comparison video coming soon.

@skeptrune I don't miss writing code by hand

realizing that i too haven't written code in months kind of just makes me want to cry

@jxnlco Threads!!!!! (steal from Pi)

If you're using codex desktop app today, what features do you feel like are still missing? Let me know and I’ll summarize all the feedback and share internally.

@Masterji_UPWale bs. yahoo mail was created 29 years ago.

Anyone else has a yahoo id ???

@zicohacks @antigravity This is intersting. I like solving problems, architecting solutions, etc, but typing the code itself? Not so much. I'm having quite some find designing systems togethers with AI.

@dimaip @antigravity bcz I love coding

Introducing Antigravity 2.0, a new standalone desktop application that delivers fully on that original glimpse of a truly agent-optimized experience. Rebuilt from the ground up with multi-agent teams, scheduled tasks, native voice and one-click integration with other Google products. Learn how to get started with Antigravity 2.0 👇

@gdb not much different. before codex it was claude code.

trying to remember what it was like to code before codex

@thepanta82 @vanjajaja1 yup. works for me. I'm quite efficient at helplessly yelling, much more so than at debugging complex stuff

@vanjajaja1 I guess it can, but that doesn't translate into your own understanding of what happened and how the system works. It just continues a pattern of you helplessly yelling via prompts and hoping things are magically resolved. (I wasn't clear enough in the OP)

Knowing how to yell at an agent isn't a replacement to knowing how to attach a debugger and inspect memory state when things go wrong

@thepanta82 Can't your agent do that?

@zicohacks @antigravity write code yourself? wdym?

@antigravity Antigravity 2.0 just removed its VSCode-style editor. Cursor did it first, now Antigravity. At this rate, where am I supposed to actually write code myself? 😂

@ryanflorence And all of the above gave great career opportunities

We loved social media, now we realize it's destroyed our relationships Loved smartphones. Now we realized they destroyed our attention But both are so ingrained into our lives we can't get rid of them, it's too late. AI shows up and we recognize immediately we don't want it

@ChShersh @kenwheeler Let's not forget that eventually it was all about tabs vs spaces. And obviously the correct answer is tabs.

@kenwheeler Don’t forget arguing about YAML vs TOML vs JSON and vs a dozen other config formats

2019 was technological masturbation. you cared more about state management libraries than the value the app they were being used for was providing. leaning into product and outcomes is infinitely more fulfilling than that bikeshedding bullshit.

@gail35zx @AndrewHart @darrenpjones Paying more attracts more people to choose from. It’s not the only factor, but a pretty strong one

@dimaip @AndrewHart @darrenpjones It’s like when some argue, if you pay more ‘they will work harder’! Someone, somewhere make it make sense. e.g. doesn’t explain why so many third raters assume huge responsibility + how can you work harder just for higher rate/perks - impossible!

I’m looking for a new personal assistant. Based in the House of Commons, you’ll help me with my diary, including supporting my evening meetings on Mondays and Tuesdays. This role sits at the centre of my parliamentary, constituency, government and special advisor teams helping me to be as effective as possible. Deadline: 29th May. More details 👇 w4mpjobs.org/JobDetails.asp…

@gail35zx @AndrewHart @darrenpjones the correlation is pretty strong, yes

@AndrewHart @darrenpjones Does higher pay always attract ‘the best’?

I have two NVIDIA DGX Sparks stacked in my office. They've been sitting there for a month. Here's my honest take. Open source AI is never going to compare to frontier models. Running quantized Kimi K2.6 and GLM 5.1 locally is cool. But practical? No. Not even close. I run all my Hermes agents on GPT 5.5 through my ChatGPT Pro subscription. Practically free. GPT 5.5 is the intelligent model in the world. Why would I route serious tasks to a watered down local model? If you need fast and accurate, you're not using local inference. You're using GPT 5.5 or Claude Opus 4.7. I'm not saying this to rage bait. I genuinely want to know. Why would anyone serious about vibe coding and AI agents use a local model when frontier is this far ahead?