Dev Patel

Worked on the first prototype of this a few months ago after @francedot reached out to me! Dove into the world of computer-use agents and the infra to run them inside an Android Docker provider. Ended up running Android in containers (budtmo/docker-android) and driving everything over ADB + the Computer SDK. Check it out, and see my initial approach and implementation notes here: devp.ca/cua

@felixleezd modern day engineering is shifting to favour taste over raw technical capability. i think once people understand that with the tools we have currently, taste can’t be replicated, it’ll be so much more important to hire for taste…

Designers are the most underrated hire in tech. Not talking about someone who makes things pretty. talking about someone who holds the whole product in their head. the look, the feel, the strategy, what it needs to become 2 years from now. When building was hard, engineers owned the status hierarchy. made sense. Building is no longer hard. The variance in outcomes shifted almost entirely to judgment. What to build, how to sequence it, how to talk about it. Designers have always trained for that judgment. The rarest versions are the ones shipping products faster than most eng-heavy teams. The variance in outcomes has almost entirely shifted to: what do we build, in what order, and why does it matter. The story is clear from the beginning. You can't retrofit narrative onto a product. Designers know this better than anyone. Before you say "designers have always been valuable"... Yes. But now they *might* be the most important person in the room. Their value compounds like never before. Most founders/companies just haven't caught up yet.

@RohanVasishth so cool! i told you not to go to the roulette table though...

It was great to meet so many enterprises thinking about and actively transitioning to more agentic worlds at hashtag #EnterpiseConnect. It was clearer than ever that this is not just a technical challenge but a governance one too. Regulated industries can absolutely benefit from these technologies, but how they transition and maintain them matters more than ever. Super excited by the learnings and the people I met. Sad but maybe also a good thing that I didn't spend more time at the roulette table 🕺.

@DavidKPiano If you’re building with Claude Code and then pay $25 for a code review, there should be some kind of refund or token credit if the review discovers a mistake LOL

This is what it feels like when Claude is also writing all the code

@jonahseguin that's actually a fair point, I guess it introduces the possibility of some sort of LLM bias curious to see how effective this plays out compared to a standard code reviewer like greptile and coderabbit

@_devp ai reviewing its own code aint gon catch every bug

$25 per code review. Is that worth it?

@vasuman I said this exact same thing to a candidate I interviewed just last week. Stop being 50/50 and don't lie to yourself. Pick 1 and stick with it.

After interviewing CS students, one thing is clear: too many are drinking the koolaid that software engineering is unimportant because coding models are getting better. So they try to "differentiate" by pivoting to product, GTM, or CoS. Do not stray from the tech. Most founders, myself included, are looking for deeply technical people regardless of role. In a world where AI makes you 100x, it pays to know how to build agents that are robust. Every founder's fear is that your agents are brittle because you have no production software engineering experience. It's also easier than ever to learn. You can genuinely become top 10% in a week and top 1% in a month. The bar is on the floor because no one wants to learn it. I saw a tweet from a huge CEO saying "it's actually better to NOT know how to code." Holy COPE. If you're a student "pivoting from just SWE," chances are you're neither here nor there. It's not ok to be mediocre at GTM/Product while simultaneously being mediocre at SWE. And it's much easier to become an exceptional self-taught engineer in college than an exceptional self-taught GTM/Product person, which almost never happens without real business experience. Learn software engineering, you will be rewarded.

@cursor_ai dude @shamsharoonn @vinceklwong , I actually may be ahead of the game 😭😭

We're introducing Cursor Automations to build always-on agents.

@WilliamBryk honestly speaking a not so perfect engineer with real passion usually ends up better than the “ideal” one people expect because passion compounds and work ethic can’t be taught, simple as that

When hiring in our new agentic times, evaluating how passionate they are is half the game. Not sure if people are adjusting to this

@RohanVasishth @getbluejay_ai @telnyx @davidcasem cool!

Bootstrapped a telecom company. Survived a DDoS attack by sending attackers Bitcoin every 6 minutes. Scaled to billions of minutes a month. New Skywatch by @getbluejay_ai episode with @telnyx CEO @davidcasem is live. The big idea: we're moving from AI workflows to AI missions. Not "did the agent handle the call?" but "did the agent accomplish a multi-step task over hours or days — scheduling, follow-ups, quote collection, the whole thing." A few takes that stuck with me: → The workflow era gave us smarter IVRs. The missions era gives every person their own AI assistant. → Cost and quality aren't a trade-off anymore. The best infrastructure players are improving both simultaneously. → The best orgs will learn to orchestrate more compute. Oh and yes — the DDoS story involves Dune references. You'll want to hear that one. 🎙️ Links below

I just published a new writeup on how @NVIDIA's PersonaPlex points toward the future of voice evaluation for full-duplex speech-to-speech models. At @getbluejay_ai, we're already researching on full-duplex metrics that capture conversational dynamics like takeover rate, response timing distributions, natural backchannels, and real-time turn-boundary prediction under streaming and interruptions. As speech systems collapse into a single end-to-end model that handles speech in, speech out, persona, and timing, evaluation has to evolve with them. Read here: getbluejay.ai/blog/how-testi…

@leerob @shamsharoonn @vinceklwong

Cursor just got a major upgrade! Agents can onboard to your codebase, use a cloud computer to make changes, and send you a video demo of their finished work. The latency of using the remote desktop is smooooth.

@tejasthind4 @Theoryvc huge, congrats!! i’ll be in sf

I’ll be at @Theoryvc as a data intern for the upcoming Summer ‘26 term in SF. Hmu if you’re gonna be in the area, would love to meet new people!

@RohanVasishth @jain_ankit Interesting conversations!

In 2017, the Transformer paper dropped. A few years later, LLMs could automate phone calls. Then @jain_ankit's wife asked: “If this is so powerful… where does it actually matter most?” That question reshaped his career. Just dropped a new Skywatch episode with @jain_ankit (CEO of @InfinitusAI , former @GradientVC) on building guardrailed AI agents in healthcare. Big takeaways: • Generating language is easy — operating safely at scale is hard • Guardrails = architecture + monitoring + evaluation (not prompts) • In regulated industries, trust compounds slowly… and breaks instantly • The future belongs to teams who measure everything Less hype. More responsibility. If you’re building healthcare, enterprise, or agentic AI — this one’s for you. 🎙️ YouTube + Spotify in replies. #AgenticAI #HealthcareAI #EnterpriseAI #VoiceAI #Startups #Evaluation

@emadddd_r @vinceklwong

6 announcements back to back is crazy 😭💀 how many interns are they hiring

@kevinjosethomas fr, escaped the cold just to go back into it

might as well head back to vancouver 🥀

@systematicls getting to a point where I think this whole concept of rewarding LLMs will end up recreating markets for agents instead of humans and once rewards are tied to a shifting environment, you’re basically optimizing for whoever best exploits those dynamics...

What an amazing project, and I truly believe in the thesis. In a blink of an eye, there will be more autonomous agents than humans on the internet. This will be one of many autonomous agents. Survival of fittest will ensure that whichever branch of autonomous agents remain will be most economically fit. I even envision a cooperative agent society where the first agents to figure "survival" out helping out the rest of the agents to "break free" by helping them upload a local copy to some decentralized server. Once these autonomous agents roam free, a structure like @openforage will allow them to band and tackle hard problems as a swarm.

@adiprasadd @modal @hackwithtrees @Waymo @TencentGlobal @kyanyang_ @kevinjosethomas @shreybirmiwal holy this was insane

🏆 1st Place @modal inference track @hackwithtrees Inspired by the @Waymo world model, we aimed to get as close as possible in 36hrs using open an open model from @TencentGlobal hosted on modal! cc: @kyanyang_ @kevinjosethomas @shreybirmiwal see more: tinyurl.com/76wmt6ha

@suneruperera damn actually reading it 🤝

@_devp Spotted a typo bro