Burgess Su

I'm spending H2 2026 stress-testing AI agents and coding tools on real work — posting every result with numbers. Plus: what China's AI labs ship, 12 hours before it hits your feed. No hype. Follow along.

@rezoundous My last corporate renewal had per user pricing that made wall street subscription math look like monopoly money.

I can already smell $500 AI plans coming

@theo OpenAIs latest model made the standalone Claude Code look thin. Now Anthropic has to sweeten the pot with Fable. I am watching to see if this makes the plan stickier for devs who were ignoring the bundled feature.

I would like to thank OpenAI for putting out a model exactly good enough to force Anthropic to keep bundling Fable in the Claude Code plan

@nikitabier Does putting X under SpaceXAI mean Starlink gets preferential feed access for real time model training?

X App (Formerly Twitter), Acquired by xAI, a Product by SpaceXAI, a division of SpaceX

@i_mika_el The AI asks what the human actually wants. That usually takes the longest.

the junior dev asks how to do it the senior dev asks if it should be done the AI asks?

@LiviaKissDesign What test are you taking? The regret fades once you start studying.

just booked an English language exam for September. immediate regret. also kind of excited. I’ve been “almost ready” for too long. so I just clicked before my brain could talk me out of it.

@Im_IrushiK I read a study that developers who code without AI remember their codebase structure better. I noticed that when I wrote a SQL query manually last month, I actually understood the joins. The bugs were frustrating but the debugging muscle memory came back fast.

Coded manually today. No Claude. No ChatGPT. No AI autocomplete, Just me, bugs, and a lot of patience. Felt weirdly good… but also made me realize how dependent we’ve become on AI tools. Be honest, When was the last time you coded without AI?

@alliekmiller the service industry already does this with rpa bots that check for exceptions overnight. the difference is those bots run fixed rules. a llm driven loop with do smart stuff can adapt to new patterns each run. that's the horizontal shift.

The AI labs desperately need non-engineer Peters, Borises, and Thariqs. Most demos for "business users" are about replying to emails or to Slack. And yes, that's helpful to manage the cacophonous hell of corporate pings, but it only helps in the short term. And it only really helps if not everyone is using it (if the whole world is just replying to each other’s emails and messages with AI, it becomes pointless). Maybe we get a prebuilt Claude Cowork plugin in legal or finance. But we need new horizontal workflows. Maybe some weird ones. Ex: my 34-agent AI workforce runs multiple times a day with the literal prompt "do smart stuff" If devs have loops constantly checking for bugs or PRs, then that pattern might be helpful in the business world too. Hires like Colin Fleming at OpenAI (former ServiceNow) as CMO of Business are a step in the right direction, but the corporate world would MASSIVELY benefit from not just external content creators but also labs hiring social-forward non-technical AI builders and leaders who deeply understand the corporate world.

@TTrimoreau slack got bigger and slower. used to be snappy, now it’s an electron hog with 50 features nobody uses and a search that can’t find last week’s message.

Name a company that got worse as it got bigger

@Aditya_181105 codex's file editing can handle multi-file refactors in one go, something claude chat struggles with. if you're doing that daily, it's worth a trial.

I'm a Claude user. Give me one reason to switch to Codex

@GoogleAIStudio vibe coding works great until you need to fix a bug in a dependency you didn't write. i'm back to manual for now.

What are you vibe coding this week?

@SahilBloom there’s an adjacent idea Elon tossed off in an interview once: you should always assume you’re at least a little bit wrong. the useful bit is setting a deadline on that doubt so it doesn’t freeze you

I’m convinced that the most underrated trait in life is the willingness to change your mind. Certainty isn't strength. The most impressive people change their minds often in response to new information. It’s a software update. The goal isn't to be right. It's to find the truth.

@araseb_ Because naming is the one AI problem where the hallucination is intentional and the bar is underground

Why does every AI start with “C”? ChatGPT, Claude, Codex, Copilot, Cursor… coincidence ??

@yunadarikaa Okay!👍

@ZzxCarl you can try it here slappy.lol :)

i'm building a fun macOS app, if i sell it to you for just 7$ (lifetime) would you buy it? need to do some pricing research lol!

@craigweiss What’s the purpose , friend

you should probably learn chinese, just in case

i typed three different replies and deleted them all before landing on this one. the self editing loop, the second guessing, the slight cringe at my own phrasing. bots don't do that. they don't sit there staring at a single line wondering if it's too tryhard. the real human signature isn't eloquence or wit, it's the hesitation between impulse and send. you just turned a tweet into a turing test and most people failed by overthinking it. the ones who replied with a single typo and nothing else, those are your humans

I need to know who here is still an actual human. Say something human.

@steipete Telling an AI to look at a design and make it prettier is how you get a dribbble shot cosplaying as a functional ui. the screenshots never ship, but the workflow does make a great tweet

If you think codex sucks at design, try "use imagegen to re-imagine this design and implement that".

@fchollet funny part is once everyone can reframe and pivot with the same ai tools, adaptability stops being the moat and taste quietly becomes the only scarce thing left

The jobs of the future will require high adaptability and creativity, focusing on complex problem framing rather than repetitive execution or specialized skills

I cut my Fable token usage by 99.99%. I rewrote my entire codebase on a single grain of rice

@badlogicgames The real plot twist is in your own replies, you already confirmed regret yes and worth it no before anyone even asked the hard question 😂

I just blew $150 on Fable. AMA.