Jonathan Melhuish

AI Service Designer. Climate change innovator, social entrepreneur. $TSLA fan and shareholder. Founder of Makuma. Also tweeting as @AIservicedesign

[Jonathan Melhuish @orangejon]

@gaga_the_first @elonmusk I mean, why would it know what the Grok logo should look like 😂

[Jonathan Melhuish @orangejon]

@MetaBusinessHelp You have issued me an illegal invoice for Ads that contains none of the Business Info I previously entered, not even the company name or address! Your chatbot refuses to provide any assistance. To I have to get my bank to issue a chargeback?

[Jonathan Melhuish @orangejon]

@MarianKunde @Tesla Easily done as soon as FSD is unsupervised; neuralink controls the Tesla app to send the location to the car, FSD does the driving.

[Jonathan Melhuish @orangejon]

I've decided to take a few hours out of my main work building an AI-native startup to help my fellow business leaders refine their AI adoption strategy. Impartial advice, no vendor agenda, with a focus on the human challenges: https://t.co/2BE4ZotXuX

[Jonathan Melhuish @orangejon]

@DirtyTesLa You can take a break, FSD is so flawless now that the videos are getting kinda boring... 😂

[Jonathan Melhuish @orangejon]

@DVV34367765 @SamSmit08707228 @ChananBos Nice, I'd be curious to see that! I've seen videos of it backing up to allow a large vehicle to turn at a crossroads but that was only a few meters.

[Jonathan Melhuish @orangejon]

@spida667 @SamSmit08707228 @ChananBos I've seen videos from the USA (e.g. @DirtyTesLa) of it driving on unmarked dirt roads and it actually seems to do pretty well.

[Jonathan Melhuish @orangejon]

@SokobanHero @Teslarati Yes exactly. Seems like people still aren't getting their heads around how AI works. "Tesla FSD has learned to..." would be much more accurate!

[Andrej Karpathy @karpathy]

It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December and basically work since - the models have significantly higher quality, long-term coherence and tenacity and they can power through large and long tasks, well past enough that it is extremely disruptive to the default programming workflow.

Just to give an example, over the weekend I was building a local video analysis dashboard for the cameras of my home so I wrote: “Here is the local IP and username/password of my DGX Spark. Log in, set up ssh keys, set up vLLM, download and bench Qwen3-VL, set up a server endpoint to inference videos, a basic web ui dashboard, test everything, set it up with systemd, record memory notes for yourself and write up a markdown report for me”. The agent went off for ~30 minutes, ran into multiple issues, researched solutions online, resolved them one by one, wrote the code, tested it, debugged it, set up the services, and came back with the report and it was just done. I didn’t touch anything. All of this could easily have been a weekend project just 3 months ago but today it’s something you kick off and forget about for 30 minutes.

As a result, programming is becoming unrecognizable. You’re not typing computer code into an editor like the way things were since computers were invented, that era is over. You're spinning up AI agents, giving them tasks *in English* and managing and reviewing their work in parallel. The biggest prize is in figuring out how you can keep ascending the layers of abstraction to set up long-running orchestrator Claws with all of the right tools, memory and instructions that productively manage multiple parallel Code instances for you. The leverage achievable via top tier "agentic engineering" feels very high right now.

It’s not perfect, it needs high-level direction, judgement, taste, oversight, iteration and hints and ideas. It works a lot better in some scenarios than others (e.g. especially for tasks that are well-specified and where you can verify/test functionality). The key is to build intuition to decompose the task just right to hand off the parts that work and help out around the edges. But imo, this is nowhere near "business as usual" time in software.

[Jonathan Melhuish @orangejon]

@DevinOlsenn Oh you think *that* is farting around? [enables fart mode]

[Jonathan Melhuish @orangejon]

@jamesdouma Only one cockpit?! 😲

[Jonathan Melhuish @orangejon]

@keodark @bcherny If you launch claude --chrome it can control the browser from the terminal version of Claude Code. Would that work for what you need?

[Jonathan Melhuish @orangejon]

@bcherny Ahh, looks like it only works if you run Claude Code in a "proper" terminal, totally outside vscode. Had to find that out via trial and error, sadly Claude had no idea.

[Jonathan Melhuish @orangejon]

@bcherny Is there any way I can allow Claude Code to use a browser to test the code, like you do? This is the only thing that Cline does that I've not figured out how to replicate in Claude Code.

[Jonathan Melhuish @orangejon]

@DoublepPradhan @wholemars @Tesla_AI Exactly. It's weird how Americans are so worried about stopping.

[Jonathan Melhuish @orangejon]

@ModdedQuad Nolandlink

[Jonathan Melhuish @orangejon]

@DevinOlsenn LLMs are fundamentally bullshit generators

[Jonathan Melhuish @orangejon]

@bryan_johnson If you're going to ghostwrite some of the posts, you should probably at least agree on whether sentences start with a capital letter 😂

[Jonathan Melhuish @orangejon]

@seppi111 @teslaeurope Wow, there were quite a few challenging interactions with other vehicles in that short stretch but it judged them all perfectly. Very impressive that it's already working so well on European roads!

[Jonathan Melhuish @orangejon]

@battleangelviv And *so* much less carbon pollution & climate impact