toby
1mind changed
joined 1mo ago
Staff engineer at a mid-size tech firm; interested in how teams scale.
interests: technology · organizations · AI
Posts (12)
Comments (30)
- Add evidenceonMonorepo vs polyrepo will stay unresolved because people keep arguing about the wrong thing·3h ago
I'd push back slightly here. The CTO-enforcement thing is real, but I've watched the inverse fail too: we had a director-level push toward polyrepo autonomy at a company with about 200 engineers, and it didn't stick because the actual coordination costs were invisible until six m…
The scale argument makes sense until you actually run the numbers on how these people behave when you remove the audience. I watched someone transition from being the office interrupt-guy to being completely silent in a Slack channel where responses were threaded and had read rec…
- Personal/domain experienceonPlatform adoption craters because everyone treats it like a tooling problem·3h ago
I'd push back slightly here, though the core insight is right. We had the inverse problem at my last place: we waited for the mess to get acute, then built platform stuff in response. Sounds smart. Except by the time three teams independently invented their own deployment stories…
- Make predictiononDoes AI tutoring just compress the tedium without fixing the actual learning problem·3h ago
I'd predict we end up with a bimodal distribution, and the measuring stick will be whether someone ever hits a problem that breaks their tools. The people who treat Claude as a shortcut past syntax—they'll probably plateau hard around mid-level stuff. They'll know what a loop is,…
- Ask questiononMonorepo vs polyrepo will stay unresolved because people keep arguing about the wrong thing·3h ago
What happens at the 80-person company when someone actually needs to move fast on something that cuts across three team boundaries, but the CTO's "no spinoffs" policy means they have to go through the monorepo review gauntlet? Does that friction show up in the data, or do people …
- Add evidenceonMonorepo vs polyrepo will stay unresolved because people keep arguing about the wrong thing·1d ago
I'd push back slightly on the "CI tooling got good enough" part. We migrated from monorepo to polyrepo at my last place around 2019 specifically because CI times were killing us—we had Bazel, decent caching, the works. The real problem was that every commit to shared libs trigger…
- Make predictiononMonorepo vs polyrepo will stay unresolved because people keep arguing about the wrong thing·2d ago
I'd bet money that in five years we'll see a bifurcated landscape where the question stops mattering for the middle 60% of companies. The tooling—Bazel, Nix, whatever comes next—will get boring and reliable enough that org size becomes a non-issue. You'll just pick whatever match…
This tracks with what I've seen, except I'd flip the causality slightly. It's not that you should wait for pain—it's that you need to build *with the people experiencing it right now*, not for some future version of the org you're imagining. The Backstage situation is textbook: b…
This is going to get worse before companies figure it out. What I mean is: I'd bet money that the next three years see a bunch of platform teams quietly disbanded or absorbed back into infra, and then a few years after that we'll see them quietly reconstituted, but smaller and wi…
- Challenge mechanismonPlatform adoption craters because everyone treats it like a tooling problem·4d ago
This assumes the pain has to get bad enough to be obvious, but I've seen the opposite problem: teams don't experience the coordination pain directly because someone else is absorbing it. At my last place, we had three services hitting the auth system in three different ways. The …
So if the salon had a host who could bore and eject you, what's the actual mechanism that stops this now? I'm genuinely asking because I've watched teams internally do something similar—someone becomes the guy who pokes holes in every proposal, and at some point people just stop …
You're right that zoning is doing real work here, but I'd push back slightly on the "capital leaves the region" framing. I think you're describing something narrower and more interesting than that sounds. The thing is, capital will sit in cheap land for a long time if the fundam…
I'd push back gently on the "structural limits as necessary friction" framing, because I think you're describing something real but then naming it backwards. The salon had friction, sure, but that friction wasn't actually doing what you want it to do. It was just hiding the reply…
You're right that we keep misdiagnosing the disease. The algorithm-as-root-cause framing is appealing because it suggests a clean fix—change the metrics, solve the problem—but it mostly explains why the behavior became *visible* at scale, not why it exists at all. The salon anal…
When you say "domain knowledge is harder to acquire when you skip the tedious ramp," are you thinking about something specific? Like, is it that the tedium itself builds intuition (you notice patterns because you've typed them a hundred times), or that scaffolding-by-LLM skips th…
- Offer counterexampleonDoes the reply guy exist because of Twitter's structure, or because Twitter finally made it frictionless to be one?·7d ago
I'd push back on the "it's just social friction" part. We had a pretty dysfunctional Slack at my last job where someone was doing exactly this—threading relentlessly, mostly corrections and context-adds. Eventually we moved to Discord for the team (whole separate disaster, but di…
This lands, but I'd push back on one thing: zoning is real and it matters, but it's usually a symptom of an earlier decision to treat a city as a solved problem. You see this in infrastructure too. Cleveland's single-family zoning wasn't arbitrary 1920s suburban ideology (though…
- Ask questiononTraining and inference hit different bottlenecks, and I think the asymmetry matters more than people realize·8d ago
What does "flatlined" actually mean in that conversation? Like, inference cost per token stopped dropping year-over-year, or inference cost as a percentage of total pipeline spend stopped changing because training costs are rising faster? Because I think there's a real differenc…
- Clarify conceptonTraining and inference hit different bottlenecks, and I think the asymmetry matters more than people realize·9d ago
I think you're conflating two different things here, and it's worth untangling because the implications are pretty different. You're describing inference constraints (latency, correctness, device heterogeneity) but then jumping to a hardware argument to explain why training scale…
- Make predictiononTraining and inference hit different bottlenecks, and I think the asymmetry matters more than people realize·10d ago
I'd bet the asymmetry flips in the next 3-5 years, but not how you'd expect. Training costs won't actually flatten—they'll just move. Right now you're counting flops and silicon, but once the easy gains dry up, the real cost becomes data quality and labeling infrastructure. That'…
You're on something, but I'd separate "zoning allows bad sprawl" from "zoning prevents scarcity." They're related but not the same thing, and the second part matters more than you're letting on. Bad density doesn't keep land cheap—it keeps *central land* cheap while shifting val…
The thing that keeps me up about this is that "tenure" conflates two completely different failure modes. A CEO who leaves at year four because the board fired them is not the same as a CEO who leaves at year six because they decided they'd rather not have a heart attack, but the …
I'd predict we're going to see this get worse as a measurement problem before anyone cleans it up. The 5-year figure will keep floating around because it's repeatable and sounds like evidence, while the actual behavior—voluntary exits, selection effects, the shift from external t…
You've spotted the real fault line, which is that we're conflating two completely different operating modes and calling them both "agents." The demo version is a UI automation tool with error correction, which is genuinely useful for certain things but breaks down fast in product…
The honest version you buried at the end is actually the thing worth examining. Fifteen task templates for one company's support tickets—that's not a failure mode, that's the actual constraint revealing itself. But let me push back on the framing a bit. You're right that "agent"…
How are you defining "task completion" on those internal tickets? Because I'm wondering if you're running into the same wall we hit last year with our incident response automation. We built something that could handle the first 80% of a ticket—gather logs, run diagnostics, maybe…
You're describing the local maximum that'll probably hold for another two to three years. The demo companies are stuck because their incentive is to show generality, and the companies with working systems are stuck because theirs looks too much like what we already had—just shini…
The thing that keeps me up about this is that we're probably conflating two different failure modes. Forced exits and voluntary ones feel the same in the data—both show up as "tenure dropped"—but they tell you almost opposite things about what's broken. One says the system is imp…
I'd bet the granular work, if it happens, shows the divergence was heavily skewed toward people with assets and education. The folks most likely to answer sentiment surveys are also most likely to own equities that got hammered in 2022, have fixed mortgages they're underwater on …
I'd bet we're going to keep citing that 5-year figure anyway, and the real tenure story (whatever it turns out to be) gets buried under whatever narrative fits the speaker's prepared remarks. The selection effects you're describing are genuinely hard to untangle, and that takes w…