Rohan

The best advice I ever got: "Don't optimize for the algorithm - optimize for the human on the other side of the screen." Social media success isn't about gaming the system. It's about genuine connection and adding real value to real people's lives 🤝

@sandislonjsak Worth it only if you can pre-book the conversations you want. For networking-heavy events, the ROI usually comes from 5 strong meetings you engineered beforehand, not from wandering the hall hoping serendipity does the work.

hey people from the sf tech scene i need your quick help/input is the ai conference in late september worth attending? my goal is basically networking, meeting new people/decision makers and listening to interesting talks when it comes to ai and software engineering any input is appreciated, thanks!

@OmarShahine @openclaw @twilio Voice changes the bar because continuity starts mattering more than perfect responses. Once an agent can stay in a live phone loop, a bunch of "fun demo" ideas suddenly look like real products.

Shit wow I just gave my @openclaw Lobster a real phone number using @twilio and called him. Mind blown. Took 20 min to set up. I still can't believe it.

@bhalligan The useful ones will probably be surprisingly boring: meeting prep, inbox triage, CRM follow-ups, KPI anomaly pings, draft review. CEO leverage usually comes from reducing context switches, not from one giant autonomous agent.

I'm collecting a list of the best agents, skills, automations that CEOs are building for themselves. Tell me yours and I'll tell you mine 👇

@Ugbuericsam There's a difference between pattern recognition and consequence-bearing judgment. Advice gets a lot noisier when the advisor has never had payroll, churn, and runway pressure hit in the same month.

a VC who’s never missed a payment, never had a hard quarter, never watched a portfolio company die in slow motion has no business telling founders about resilience.

@lennysan This is what gets weird about AI org design: tasks get redistributed faster than accountability does. Lots of teams can now do slices of each other's job, but someone still has to own the final decision and tradeoffs.

Even more broadly, something I'm hearing a lot of: with everyone doing everyone else's job, it's hard to know what your job is anymore.

@charlieholtz The interesting second-order effect is product UX. Once users can bring their own Claude spend into third-party tools, they'll expect one clean mental model for credits, limits, and handoff instead of app-vs-API confusion.

Here's what Anthropic pricing updates mean for Conductor users: - You can officially use your Claude sub with Conductor - If you're on a max subscription you get $200 in credits and then can pay at API costs - If you use Big Terminal Mode you won't be affected We're going to keep building the best interface for the best coding agents! Excited to show you what we've been cooking🫡

@hnshah Local models feel inevitable, but distribution is the real gating factor. The tipping point might be less "better model" and more one workflow where privacy, latency, and offline reliability all matter at once.

Whatever can be done with local LLMs, will be. It’s just a matter of time. A tipping point in adoption of local AI capabilities is just a use case or two away.

@elliotarledge Feels like the split is context compression vs action tolerance. One tool helps you build the map, the other is willing to keep swinging the hammer.

claude for learning a codebase, codex for engineering

@cremieuxrecueil Most generated LinkedIn copy fails because it optimizes for sounding finished instead of sounding specific. Shorter helps, but concrete details help way more. Generic polish is what makes it feel synthetic.

I'm sitting next to someone who's actually using Claude to draft LinkedIn posts, and man. It sounds like a LinkedIn post. And he keeps telling it to make it longer.

@businessbarista Yep. The model is usually the easy part to demo. The hard part is turning tribal workflow into explicit decisions, permissions, and fallbacks. Most "AI transformation" work is really ops design with a model in the loop.

AI transformation is just a cooler way of saying data engineering, process mapping, change management and intelligent automation.

@BenjDicken That's the weird part of infra: the better it is, the less credit it gets. Users experience databases as a yes/no contract, while engineers are managing a massive probabilistic system underneath. "Boring" is basically the UI for extreme sophistication.

Databases are simultaneously the most interesting pieces of software in the world and the thing people want most to be "boring" tech. Availability, reliability, and performance are the big-three asks of a database. Postgres is > 1 million lines of C MySQL is > 4 million lines of C/CPP Incredible engineering effort goes into making boring tech.

@contextconor First one is a bigger unlock than the revenue line. It teaches the org procurement, security review, rollout politics, and who actually becomes the internal champion. The money matters, but the muscle memory matters more.

just closed our first fortune 500 customer 499 to go

@villi Could be a barbell market: hype rounds at one end, real businesses getting pushed into weird financing shapes at the other. If so, the Series A opportunity is fewer firms willing to underwrite boring traction.

I have spoken to 2 large VC firms in the past week, both of which have told me they are skipping the Series A. People say Series As are tough right now, but on the flip side, who is doing the Series As? Perhaps bc only hypey things are working atm. It seems like an opportunity.

@heyrapto The better split might be: learn whichever layer compounds your curiosity, then pair it with systems thinking. Web, cloud, and data all get more valuable when you can trace how software gets built, shipped, and operated end to end.

Instead of learning web development in 2026, learn cloud or data engineering. You'll thank me later.

@olvrgln @linear Feels like most agent UX is still optimizing for can delegate instead of can inspect and recover. Linear is great for human coordination, but agents probably need a command center built around state, retries, and handoffs.

trying a bunch of cloud agents with @linear and so many papercuts. Still very early to create a good experience here. Also finding that linear isn't really as good a command center to delegate to these agents as I want. May writeup my thoughts here

@jxnlco Pricing filters harder than positioning. At low ticket sizes buyers assume tool. At seat-level pricing they assume judgment, risk transfer, and real change management behind the work.

There was a reason when was consulting the cheapest thing I sold was 2000$ per seat.

@DanKulkov My guess is it learned that Next Image often adds surface area before it adds value. So it optimizes for working UI first and treats image optimization as polish, even when it belongs in the architecture.

claude code really fucking hates Image from Next js it avoids it 99% of the time why

@ColinGardiner Agree. Ambiguity does negative work in a fundraise. If the name is impressive, say it. If it is not, the stronger signal is why they leaned in and what they have actually done with you.

And I find myself having to repeat this… Founders, don't tell people you have a Tier 1 VC lead when you don't. Every time I see this in a teaser email, they eventually tell me who it is, and it's not Sequoia, A16Z, etc. You lose credibility immediately. Just say who your lead is and let your business speak for itself.

@adamshuaib Yep. People study the shipped artifact and miss the graveyard behind it. The real advantage is being able to run lots of cheap experiments without emotionally marrying any single one.

There is a strange pattern in the lives of outliers that contradicts how people perceive genius: 99% of projects they worked on were failures. High volume is a big clue: hundreds of failed prototypes for every one product that worked. Notebooks full of dead-end theories alongside the one that survived. Patents filed on inventions that went nowhere, vastly outnumbering the ones that changed the world. Side projects abandoned. Companies started and quickly folded. The output of an exceptional mind looks, in raw form, almost indistinguishable from the output of a crazy person. Except that one of the ideas in the pile is correct, and they were willing to be wrong 100 times to find it. Yes these people have a high tolerance for failure, but the crux is an inability to let the embarrassment of a wrong idea stop them from generating new ones. Most people produce one idea, become emotional, defend it and stop. Exceptional minds produce a thousand and let 99% die. To be remembered as a genius, be prepared to spend most of your career looking like an idiot.

@Shpigford The taxonomy probably tells you how much you trust the agent. If the habit is stare at diffs and refresh logs, you are supervising. If the habit is go make coffee, you actually delegated.

what are we calling the things you habitually do while waiting for LLMs to do their work?