Wondering why your maker-turned-manager suddenly seems distracted in meetings? Maybe they're addicted to coding agents! In this episode of Lightcone, Calvin French-Owen — a co-founder of Segment and former engineer on OpenAI's Codex team — joins us to talk about why coding agents suddenly feel so powerful, the differences... between Codex, Claude Code, and Cursor, and what the future of work will look like. 0:00 Intro 1:15 Garry can’t stop using Claude Code! 4:00 Contrast with IDE’s, context-splitting 6:23 Distribution models, top down vs bottom up 9:11 Licensing and optimization 12:28 Tips on becoming a top 1% user of coding agents 17:36 When can the agents work 24-48 hour running jobs on their own? 21:34 Can the agent teach things like architecture? 26:27 Will the next generation have even better taste and multitasking ability? 29:58 Maker vs manager schedules 31:36 How would Calvin build Segment now? 35:52 The importance of testing 38:52 The Claude bots are talking (to each other!) 40:10 Examples of complex issues, how will the tools evolve 43:00 Outroshow more

Tune in:

@calvinfo Thank you so much~♡

@calvinfo Do coding agents ultimately become so adept at writing code that the human touch becomes harder to distinguish, potentially altering the dynamic between makers and managers in a way that's beneficial for productivity, but also erodes the value of human intuition?

@calvinfo lmao this is too real 😂 I've caught myself doing exactly this - in a meeting but my mind is just thinking "wait I could automate this with cursor" or "the agent could probably handle this task"... coding agents are genuinely addictive fr

@calvinfo Coding agents are pulling managers back into maker mode whether they planned it or not. The real question is whether that's a feature or a bug for the org. Probably both.

@calvinfo wait managers are actually coding now? or is this about them constantly checking what devs are building lol

@calvinfo Does the ease of use of coding agents outweigh the long term risk of dependency on these tools for developers.

@calvinfo bro, sounds like they're deep in the code trenches lol

@calvinfo The real question isn't why managers get addicted to coding agents. It's why they stopped coding in the first place. Management was the tax you paid for scale. Agents are removing the tax. The best founders were always builders pretending to be managers.

@calvinfo this was such a good episode. yc videos really give you the insights.

@calvinfo The dirty secret: once you've felt the dopamine of shipping a feature in 20 minutes that would've taken a day, sitting in a meeting about Q3 OKRs physically hurts. Coding agents didn't create the addiction. They just made the relapse instantaneous.

I listened to this episode too, and it genuinely helped me lift the hood and see what’s really going on underneath. They don’t just talk about tools at the surface level. They walk through the mechanics, the failure modes, and the tradeoffs, which is exactly the kind of detail that builds real engineering intuition and taste. I also used NotebookLM to turn my notes into a few slides and redraw the key ideas in my own way. Re-sketching things helps me think. The two visuals below are the ones that stuck with me most. The first one captures something personal. After years in manager mode, I realized my coding muscles had atrophied. I was directing, reviewing, and planning, but rarely writing real code myself. Getting back into deep technical work felt like a marathon with a bad knee. Claude Code changed that. It acts like a bionic knee for the technical lead. Suddenly I could move fast again, traverse large codebases, debug complex flows, and iterate without friction. It genuinely feels like flying through the code, the same way Garry described. Not replacing thinking, but restoring velocity. The second visual is a failure mode I’ve hit recently: the so-called “context dumb zone.” As the prompt grows past a certain size, the model’s reasoning quietly deviate from our original design. I ran into this during a large re-architecture. Claude Code began to lose alignment with our core architecture principles. To fix it, I asked it to explicitly revisit the architecture, then anchored the key decisions in a file so those principles stay persistent and reloadable, and survivable during context compaction. Since doing that, the drift is much lower.

@calvinfo The era of “learn to code” is over. Now it’s: Learn to command code. Agents are the new leverage.

@calvinfo can confirm, completely addicted. went from managing a team to running 5 parallel claude code agents building fazm and I ship more in a day than I used to in a week. the dopamine hit of watching agents solve problems in real time is genuinely hard to step away from