siliconc0w

A few things:

0) big companies already are very comfortable using contracts to trust other people with their data. Maybe if they're inflexible on the 30-day requirement for fable some orgs will opt out but by-and-large it's already happening and it's not a blocker.

1) Cost will be a blocker. The level of token spend is untenable and the pareto curve is flattening. Most orgs are going to default to either using a distilled model from China or a distilled model from the model companies (e.g, Sonnet 5). It'd behoove Claude/OpenAI to offer a model router before another vendor wins that area.

2) Karp is selling his book. No one knows or cares what an 'ontology' is. From what I can tell, company's product is a tool that helps governments bomb people.

Seems like rationalization - PE firms establish regional monopolies and jack up rates. They do it with every type of service. Just accept that we're all serfs under the billionaires who extract rents over every aspect of our lives.

Google pays their oncall a % of their full-time base salary depending on the oncall tier (5 min response time vs 30 minutes).

This should probably be required - there is a different mindset and set of restrictions when you're expected to pick up a page. It also forces companies to use on-call judiciously - not every service needs a 5 min SLO.

What are the odds this is going to become another avenue for grift - magically any companies the trump family invests in are going to get access. Any companies that aren't sufficiently 'loyal' to the regime will have to wait or may never get access.

The reality is that there is no escaping management. A high level IC is pretty much just an unofficial manager. You are responsible for a large technical area - it's your job to meet with stakeholders, design the roadmap, build durable team processes to maintain velocity, mentor and identify the right work for the right people. You may not have back to back meetings but between 1:1s, stakeholders, projects, and ad-hoc fires they're likely still a majority your schedule. You are expected to lead without authority. Leadership will change priorities and reorganize the teams every six months. If you're focused and deliberate, you can maybe get can your project landed before the next reorg.

Randomly I was thinking about how to "prove" you are a distinct engineer who works at a company without revealing your identity.

I feel like labor organizers should offer employee verification as a free service to get people to sign who fear retaliation. Essentially upload your W2, get a token, sign petition with token. Or maybe just mail out a QR sticker to place in an employee-only area...

I don't see how more advanced models won't get gated to specific known KYC'd entities. Classification-style guardrails will never be sufficient. Distillation attacks too are really hard to prevent. Open-source models can have their guardrails easily stripped away so it'll be incredibly dangerous to continue to release more and more capable OSS models that can and will be used to give bad actors 100x leverage.

I had a similar experience after a recent job search and started working on a 'kube-lite' that just uses object storage for coordination and normal cloud primitives like auto-scale groups (skiff.pages.dev).

I ended up in a different non-SRE role but if you're interested in working on it, please let me know and I'd love to walk you through it.

Remember when the admin tried to kill anthropic with a ridiculous national defense supply chain order that the courts blocked - I'm not sure why anyone would believe them credible now

The market is supposed to do that. Once an opportunity is identified there is a rush to compete and margin disappears, so the people still get the new, better thing, but for much cheaper.

What can happen though is that companies figure out how to prevent meaningful competition to preserve high margins. They're worth millions for the innovation but they get to billions through anti-competitive and extractive practices.

The admin just tried to kill anthropic with a ridiculous national defense supply chain order that the courts blocked - I'm not sure why anyone would believe them credible now.

I wonder if this is specific to the animus toward anthropic or if this is the new industry wide level cap. Seems like a pretty big problem for the AI market in general, a lot this investment is predicated on better and better models.

Sadly, I'm getting a lot of forced downgrades to Opus for questions that are far removed from any security topic.

I can't really see wide adoption of local LLMs unless prices really start to climb. It makes sense to use cheaper hosted smaller models like Sonnet or even Kimi but these won't run a Kimi-class model and that is really the floor for non-toy agentic tasks. Spending 5k to avoid a $20 subscription really only makes sense for niche security reasons.

Just do one project. Ask them for an interesting or challenging problem they had, go over it at a high level from a business and technical perspective, and then dig in as deep as you can on specific parts. Ask about business impact, how was that measured, what were the tradeoffs, what made it especially difficult, what were the alternatives, what they'd do differently, etc. If they can go deeper than you can, then that is a good sign.

I use the $100/mo sub but my 30 day API cost is about $1700/mo.

It really depends how you use it, if you're using prompts to generate detailed designs, breaking those into lists of tasks, and then feeding those to multiple agents - it's really easy to burn through many thousands.

If you're being more deliberate and using a few agents at a time interactively, having it review PRs/resolve issues, automated clean-ups and performance optimization, etc it could be more like $1500.

If you're just throwing it one-off questions like a better stack-overflow that is well under a $100.

I've really gotten into /goal, if you can find something verifiable and leave it overnight - it's kinda like christmas morning to see where it landed.

I like take home projects the candidates then need to present and answer questions about. LLMs just mean you can be more ambitious here (though you should pay for their tokens).

The problem with provisional employment is that it can take quite awhile for a new hire to be productive. In a complex FAANG environment I'd wager it's about six months before they're not a net drag on the team and maybe 1-2 years until they're close to fully productive. These are complex not just technologically but organizationally and there are tons of hidden rules and micro-decisions that are the difference between a project stalling vs moving forward.

It would be hard to give up on tech as I genuinely enjoy building, watching systems come alive, figuring out the puzzles through when they break. I do like the term Neo-Amish though and definitely relate.

I do recommend people get outside activities to balance things out - just walking my dog 1-2 miles a day is like therapy for me (and a good way to get unblocked and energized with a new idea).

These are essentially sociopath screens where they expect you to memorize some STAR stories and regurgitate them on demand. And I don't mean screen out.

I recommend spending some time getting a few parts of the codebase idiomatic and then @-ing those files as exemplars. This works a lot better than trying to steer it with markdown. This works reasonably well for like FastAPI but JavaScript seems to be the worst, even with guidance and exemplars it'll prefer in-lining a bunch of garbage rather than use the APIs as directed.

Most startups don't actually make profits and nonprofits can't give equity so it's not really a favorable structure.

https://agent-flywheel.com/ (largely just the core workflow)

I agree that every so often you have to clean up a mess and the illusion breaks. Even with a super detailed spec, even with AGENTS and SKILLs specifying certain patterns or practices, even with 'fresh eyes' reviews from other agents, etc there are still these long tail of issues where I have to either hand hold the agent or just manually rework the code. Some examples:

* it cheats at verification. Even with specific instructions how to verify, it still cheats.

* generating UX(CLI tool) that is absolute garbage and inconsistent, even with specific instructions to minimize unnecessary flags, use convention over configuration ,etc.

* it absolutely will not go 'above and beyond' to solve problems - if task is hitting a permission or dependency barrier, it'll likely cheat or handwave the problem away. (gpt 5.5 xhigh)

There is maybe this hope/hubris that we can figure out just the right incantations or agent workflows to eliminate these issues - I was optimistic about this too but after trying for awhile and seeing them not only not go away but in some cases regress with newer models, I am less sure.

I've been using agent flywheel workflow which is similar. Still not completely sold - it feels a bit like using power tools to shape wood but the final product needs a lot of sanding and polishing.

I thought initially this meant that the spec wasn't detailed enough but the problem is more agent adherence and laziness.

Google has amazing potential but has consistently squandered it. Gemini CLI being killed/rebranded is yet another example of their complete lack of follow through and persistence. It wasn't a good product - it was slow, buggy, and unreliable but you have to fix it to demonstrate you can do more than launch and then kill products.

They have everything going for them - amazing technology and technologists, huge distribution and lock-in, and a giant compute advantage and the can pay for more out of cash flow rather than debt or equity. And yet it's still hard to see them not fumbling the ball.

EU equities outperformed US in 2025. The Iran war will probably shift this back to the US but launching a new poorly defined war (and arguably losing it) is also a pretty good indicator of decline.

Why would you use an infrastructure provider on top of another infrastructure provider? It adds cost and risk, it's always going to be a leaky abstraction, and it's not hard to learn how to use GCP or AWS correctly - especially with agents.

People don't really understand that non-trivial software development isn't even 50% coding. The coding step is generally the 'easiest' part and given to Junior developers. In a large org most product changes span multiple systems and human operations. Seniors and even mid-level generally spend most of their figuring out how to shape the local priorities into a new arrangement of the existing cybernetic entity and then getting buy-in on that new vision given these other teams have their own priorities.

This naturally involves a lot of tradeoffs and politics - senior engineers know to avoid adding 'weight' to their airframes and fight hard to avoid adding scope to the systems they're responsible for or divergence from their intended direction of travel. So compromises have to be struck or escalations to management to choose between priorities have to play out.

Maybe AI solves that as well but that is a lot more difficult lift.

Both OpenAI and Claude already charge Enterprise usage rates and they're still buying.

On one hand, this is insidious when targeting children.... On the other these kind of metrics are what pretty much every company tries to optimize.

We're going to get better and better at hacking the human brain - for good and evil and we're going to have to trade some free will and personal liberty to really keep the worst of it in check. The dark pattern bullshit is the easiest thing to regulate but I don't have a lot of hope for even that.