Andrew

I build AI systems, automate workflows, and test monetizable ideas in public. Sharing the lessons, experiments, and patterns I learn as I go. If you’re into AI, automation, and building online, follow along.

@svpino This is a great move. Staying under 50% context and dumping to file when it heats up is pro move. Most people fight the tool. Smart operators make the tool fight for them.

A Claude Code session becomes unusable at around 60% context. Auto-compaction kicks in a little bit over 80%, and that's when most people notice what's going on. I don't let my context get past 50%. Run /statusline in Claude Code, and it will always display how much context you are using. Once I get to 50%, I dump the entire context to a file and /clear it.

@HedgieMarkets Anthropic quietly doubling effective Claude Code costs is what I would come to expect. When the usage increases, prices follow. The edge is still building your own stack instead of depending on their billing meter. The current problem being initial cost….

🦔Anthropic quietly doubled its estimate of what Claude Code costs enterprise developers, from $6 per active day to $13, with costs below $30 per day for 90% of users. The previous estimate was based on Sonnet 3.7 as the primary model. Anthropic says the updated figures reflect Opus 4.7 now being the primary Claude Code model and that there was no pricing change. At $13 per active day, monthly costs run $150-250 per developer. Anthropic's head of growth recently acknowledged that current subscription plans weren't built for current usage levels, with engagement per subscriber described as way up. My Take Calling this a model upgrade rather than a price increase is technically accurate and practically irrelevant to any finance team looking at the line item. If the cost to do the same work doubled because the model got better, the budget impact is identical. The companies that cut engineers to fund AI subscriptions based on one cost number are now operating against a cost structure that is a different number, and every data point this year points in the same direction. This is the circular financing problem made concrete. Anthropic loses money on every token sold, raises prices as usage scales, and the startups and enterprises building on top absorb the increase with no corresponding guarantee that the productivity case they sold their boards has materialized. Uber's CTO blew through his annual AI budget before the year was half over. Goldman found inference costs approaching headcount parity. GPU spot prices up 48%. The cost of the AI bet keeps rising while the evidence that it's paying off remains, by the industry's own admission, largely unproven. I've been watching this compress all year and the direction has not changed once. Hedgie🤗

@solo_levelingx It does look like theater at times. But the compression is happening in real time. The labs making actual progress on agents and harnesses will separate from the ones just hyping the next model drop.

This "AI Bubble" is a completely hilarious meme to me We have these different "labs", all they produce is some semi-useful chatbot with different interfaces and charge a monthly subscription and then raise money by pretending the super intelligence is coming soon for years now

Exactly. Resumes are becoming background noise in AI lately A 30-minute live trial where someone ships a small agent workflow or recreates a prompt system tells you more than their last three job titles combined. The gap between “knows about AI” and “can actually drive leverage with it” is massive right now

It’s wild to see how many of the best AI startups have added work trials as a key part of the hiring process. When things are moving this fast, a candidate’s background on paper is often less relevant than what they can do with the tools in front of them.

Context hack that actually works: Feed the model your last three failed attempts before asking again. “Here’s what didn’t work and why. Now improve.” This can help turn frustration into results

@DavidSacks The key point most miss: these models don’t create new vulnerabilities, they expose the ones already sitting in the code which is a great point you made. This upgrade cycle will be massive once defenders get real access before attackers do.

It’s time to demystify Mythos. Mythos is not magic. It’s not a doomsday device. It’s the first of many models that can automate cyber tasks (just like coding). OpenAI’s GPT-5.5-cyber can now do the same. And all the frontier models (including those from China) will be there within approximately 6 months. It’s important to recognize that these models do not create vulnerabilities; they discover them. The bugs are already in the code. Using AI to discover and patch them will actually harden these systems. The leap from pre-AI cyber to post-AI cyber means that there will be a big upgrade cycle. After that, however, the market is likely to reach a new equilibrium between AI-powered cyber-offense and AI-powered cyber-defense. Obviously it’s important that cyber defenders get access before cyber attackers. That process is already underway but needs to happen quickly (see point above about Chinese models). Unlike Mythos, GPT-5.5-cyber appears not to be token constrained so it may be the first cyber model that defenders actually get to use.

@packers_owner_j @emollick At this point that feels unavoidable generally speaking…

@ShipNotHype @emollick It does get you replybot slop tho

Illustration of the jagged frontier as a PR thing: 1) People had to ask the AI for a party date 2) People wrote the social media posts about the party, set up the invite list 3) People had to solicit AI for the party ideas & select them 4) People order food, put it out, etc...

@emollick The real capability gap isn’t between models anymore. It’s between raw APIs and the native apps built around them. Codex and Claude Code prove the model performs better when the harness is purpose-built.

Increasingly, I think, we will see a gap between what you can do with frontier model APIs & what you can do with the native apps from the frontier labs (Codex, Claude Code). Models developed and trained with their native harnesses in mind have more capabilities in their harnesses

@signulll Yesterday: brainstorm 47 new ideas. 
Today: delete the 46 things that should die so you can actually move.

the ai era inverts the age old question. “what should i do” is now the lazy frame. “what should i not do” is where the alpha is.

@thejustinwelsh The old playbook said move to the big tech city. The new one says: build in public, ship daily, and connect with the sharpest people on the internet. AI collapsed geography. Your network no longer needs the same address.

@bscholl Soon we’ll need a reverse Turing Test: can this human hold a conversation without checking their phone or repeating talking points?

Feels like we are approaching a new era of AI where machines pass the Turing Test but humans don’t.

@PeterDiamandis This is the real unlock. Capital used to be the gate. Now it’s execution speed and judgment. Now $20 plus consistent output beats most seed rounds theses days.

I'll say it again: You can literally get an AI account for $20/month and start changing the world. You don't need to raise billions or millions.

@fel1de Totally! The AI gives you speed of execution. Humans give you the “dude, that’s actually stupid” that we sometimes need. Best results come from using both.

@ShipNotHype the feedback loop alone is worth more than most advisors

I build AI systems, automate workflows, and test monetizable ideas in public. Sharing the lessons, experiments, and patterns I learn as I go. If you’re into AI, automation, and building online, follow along.

@TheGeorgePu Love that implementation! This is personal leverage in action. Replace manual tracking and high-touch services with agents that handle the boring parts. You keep the judgment and the results.

Always thought calorie tracking was for gym bros. Hated it. Quit every time. Now I let AI do it. Weekly weigh-in. Waist check. Mental note on lifts. That's the whole system. Used to pay $80/session for a personal trainer. Hundreds a month. Now I pay for compute. Spend compute, not cash. That's the new personal finance. Why aren't any save-money gurus writing about this?

@iruletheworldmo I’m curious how quickly they can adapt for local LLM as well. With everything moving so fast you never know what we we’ll see the next few months!

the new gemini model is going to be well over ten trillion parameters and much more capable than the current sota. we are entering into a new and much quicker era of progress.

@megbear @bhalligan Spot on. Working code was the old golden goat so to speak. Now it’s execution at scale with agents that seemingly “disappear” into the tools. That’s where the real leverage lives.

@ShipNotHype @bhalligan I’ve been thinking about this a lot. We used to say “working code trumps theory “ I say that still holds but shifts to execution (the meaningful use at scale)

What's the smartest, fastest way you've seen a company force-multiply their people with AI? Just saw the most clever way a founder is AI-pilling their entire 300-person team. Writing it up to share, but I wonder if it can be topped...

Claude Code just charged extra because a commit mentioned “OpenClaw.” This is what happens when labs prioritize safety theater and billing over actual user experience. Vibecoding your own agents is no longer optional.

@garrytan Anthropic built the ultimate vibecoding tool then let it freak out over fictional claw references. The AI is scared of its own shadow and still wants your credit card

Anthropic shouldn’t do this and doesn’t have to Would be great if it stopped

@GG_Observatory This is fact. I’ll give it my best haha

@ShipNotHype "shipping folklore" is the best summary of how most AI projects run. the folklore-to-evidence ratio is still like 10:1 in production. glad you caught the thread. keep shipping these insights.

@PolymarketMoney $900B valuation for Anthropic is insane on paper. Either the market is completely detached or they have something massive cooking that actually changes the economics.

JUST IN: Anthropic is now considering raising a new round at a $900,000,000,000+ valuation.