Brian Turcotte

Filed under: things that are more fun to write about after the IPO prices.

The shift from chat to agents is the shift from one conversation at a time to two dozen running in parallel. Most people have just not made that mental jump yet. Once you do, going back to the old workflow will feel like using dial-up after fiber.

Developers do NOT want overly commercialized marketing fluff. They want the concise truth, whether it’s good news or bad. It’s really not that complicated, but almost no one in AI marketing is doing it.

Still thinking about that guy from #ClawConNYC who set his agent loose to self-improve for a whole month with no human check-in. If you are out there, please send a status report. I’m invested now.

I'm going to build an entire e-commerce business on stream using @kilocode, 1 hour at a time, and the first session is tonight at 7pm EST I'll be building the storefront from scratch, designing the merch live, wiring up integrations with Shopify, and using KiloClaw to automate the boring ops stuff so I can keep building. From there, I'll scale it like a real business. If you've ever wanted to see a solo e-comm build go from zero to launched, this is that! luma.com/1ajrfm2u

People got excited about running OpenClaw on a Mac mini because it looks like the movie version of AI – a physical box in your house, quietly controlling things. Cool aesthetic, bro, but it’s not actually the easiest way to use this technology. A web dashboard you can open in 60 seconds is.

The math was never going to math. $20/month for unlimited Claude runs? The physics of inference cost don't negotiate.

Built clawsweeper, which runs 50 codex in parallel around the clock, scans issues/prs deep and closes what is already implemented or what makes no sense. Closed around 4000 issues today, a few thousand are in the pipeline. (rate limits are rough) github.com/openclaw/claws…

Fair on the ROI math, $350 does beat a week of hand-writing code. The piece that comes up in DevRel conversations is that developers aren't running spreadsheet forecasts on their tools. $20 that turns into $350 without warning breaks the part of the relationship where a developer trusts the tool enough to stop thinking about it. The tool being worth the money and the bill being predictable are doing different kinds of work for the person using it.

Burning tokens isn’t cheap, so that’s no surprise. Also, the entry-level subscription for most of these tools feels more like a “free-ish trial v2.” Cursor is definitely a great tool, as are Codex and Claude Code. The best one might win this race, OR we may just keep having several strong tools, like we already do. BTW, compared to writing code by hand every week, $350 is far cheaper.

I met a guy at #ClawConNYC in March who told me he wasn’t interested in practical use cases. He wanted to give his agent max permissions, tell it to scan ClawHub and download new skills every hour, and only check in after a month. A literal AI petri dish. It has been a month. I need to find him.

fair on the Opus comparison, that's a real point. The deeper thing though is what "there are solutions" does to developer trust. BYOK and Composer exist, but every one of them is a developer doing homework to work around a pricing model that surprised them. That's the part of this that matters long-term. Tools that keep their developers close don't make them become pricing analysts to stay in the plan

@kilocode @robinebers but why are you comparing opus 3 era pricing with opus 4.7, even now, no one. No one will ever choose the old model, even if it's cheaper. If price is an issue, BYOK, use Composer, there are solution out there.

Cursor's $20 Pro plan used to get you 500 premium requests. Under the new credit system, that $20 buys roughly 45 Opus calls before your credits run out. One dev burned through $350 in a single week on overage. What used to be a working plan has become a trap door.

There is no single winner of the frontier model race. The future is multimodel and model-agnostic, with many specialized labs producing many specialized models and no monopoly at the top. Watching Anthropic vs OpenAI vs Google like it is a heavyweight title fight is watching the wrong sport!

How does open-source fit into AI? My take: it's going to become a hard requirement for making good software. We're in a gold rush right now. The Cursor/SpaceX deal makes that clear. But when literally anyone can build deeply technical things with AI, proprietary code stops being the moat. The moat is the community around the code. Who shows up. Who contributes. Who cares enough to make it better than any single company could ship. That's why I like working at Kilo. We're betting on open source as the baseline, not a nice-to-have.

The AI discourse right now is two camps yelling past each other. One side says it is a bubble and nothing fundamental is changing. The other side says it will replace everything and end society. Reality is almost certainly somewhere in the middle. It is already changing how people work, and both camps will be wrong in the same way.

Is AI actually going to change the world and no one is ready, or do I just need to go outside more?

The Kilo team retired from 7 tasks this week.

@ShaileshChotoe @kilocode Hey there! For this case, I’d recommend first checking that you’re on the latest version of the extension, and if that doesn’t resolve it, messaging our support at hi@kilocode.ai It’s not reproducible for me, so we can take a closer look there!

@kilocode the revert back to checkpoint option does not work after the update, i cant revert my code at all after each prompt 🥲

The @kilocode team reached out to me and showed me their badass new Gas Town UI inside Kilo. It's got convoy viewers, dashboards, onboarding, and other goodies. I'm pretty impressed that they were able to do it under the radar without me having to help out. I saw a few PRs come through, but it was all using the extensions, nothing in Gas Town core. Solid work by their team. Here's their blog post about it: blog.kilo.ai/p/gas-town-by-…