Pascal Brokmeier

The flood of AI will turn almost everything into an “investment decision problem”. 1 VC firm, 40.000 investment memos per month. 1 biotech 1M+ drug candidates. LLMs excel at generation not at reduction. Committees are old school Spent the last 2 years building this. Let’s talk?

@andruyeung All true but recreational drug use is also at an all time high. People pop pills more casually than ever before. There are 2 festivals for every 1 resident in some parts of Europe it feels. Is recreation generally going up and only alc going down?

Clubbing is dead and has been replaced by fitness & wellness. Ppl used to party to socialize and date but now they do things like HYROX, bathhouses, and running raves. The death of clubbing is something to be studied: — US has lost 12% of its nightclubs in the last 24 months — 25% of US adults didn’t drink at all last year — Gen Z drinks 30% less than Millennials did at the same age On the flip side: — According to Strava, the number of running clubs recorded on the platform increased 3.5x in 2025 — 72% of Gen Z go to run clubs to meet new people — Sauna and spa market: $11.8B → $22.4B by 2034 The post-alcohol economy is gonna be a massive category.

@ZackKorman Probably a luxury claim but the best companies are more interesting than any side project the individuals can come up with

Companies are like "we are spending all this money on AI but we don't know what the devs are even doing with it." Let me answer that for you: They're working on their personal side projects.

I never understood the idea of putting a middleman like an electricity provider between you and your electrons. Dynamos exist and in the age of excess calories why not put the work in and charge your own batteries?

@itsolelehmann @aisolopreneur Ofc I recognize that. In fact I got the exact comment about my partner, dealing with my city hopping and it made me re-appreciate her even more. Apologies if this offended you, I just wanted to highlight that most people couldn’t handle this and you found someone that does.

@PascalBrokmeier @aisolopreneur Have you ever entertained the thought that people differ from yourself? It’s not a decision that I have made and she has to deal with that. It’s something we decided together. Very weird comment mate.

traveling on your own: - arrive - unpack your bag (optional) traveling with a small child: - make all power outlets safe (literary the first thing our baby does it to put his fingers into outlets, oh boy) - check if any furniture can just fall over from climbing - check if anything is lying on the floor (we found everything before, from broken glass to nails) - prepare the bed so baby can't fall out (if you do co-sleeping like we do) - prepare an area where baby can play - get a highchair to feed - do all of this while keeping babys sleep rythm intact we are moving every 2 months right now to find our new place to settle but man this is stressful as fuck happy when this phase is over lol

@tushark_29 @m2jr Interesting, how does Claude make them / commands without having invoked that MCP server though? I thought MCP recently became less context polluting and it only loads tools/searches for them when needed

@m2jr For company-wide distribution a simple remote MCP server with prompts(claude makes them slash commands), resources and few tools - load_skill, load_skill_section is the simplest maintainable solution rather than asking people or automating sync. Skills are nothing but CAT command

Has anyone created an internal GitHub but for skills and prompts (not code) that can be hosted internally to a company and have non technical contributors?

@m2jr The notion pages are all in a notion DB and we use the verification tool + owners and RBAC from Notion to ensure someone ones it and not everyone can edit it but everyone can leave comments to improve it.

@m2jr We just use Notion and then create org shared skills that simply say “read this page, follow the instructions there”. Feels like the 2026 version of a C pointer 😅

@kiwicopple @supabase How’s this not the norm? It absolutely blows my mind to hear how employees walk away and loose it all or have to buy illiquid shares with a net salary that was non competitive as the shares where the “upside” replacing the 30% higher corporate job.

Supabase has raised $500M at a $10B valuation In this round we are giving @supabase employees the opportunity to cash out 25% of their vested options. We have done this in every round since inception. We do it as a “cashless transaction” so that employees don’t need to front any cash to exercise their options. This is the friendliest way we could design it until we can offer RSUs. On top of that, we give employees a 10 year exercise window: whether they stay or leave the company. The typical/default window is 3 months. IMO, equity is earned and employees shouldn't be penalized because they don't have the cash to exercise within 3 months of leaving a job (often that's the time they need the cash/certainty the most).

@harjtaggar @garrytan Now the key question: how do you recognise this during the hiring process

The best hires are the ones you can delegate outcomes to, not tasks. Good hires come back with smart questions about the next step. Great ones just get the job done.

@petergyang Symbolic link? I have skills in a global ~/Code/ai-skills and symlink those with a small script to all the global Claude codex open code etc locations. That way all these tools have my skills globally but the instructions to each repo and subfolder locally.

It's pretty annoying that Codex uses agents.md and Claude Code uses Claude.md. There should be some industry standards to this stuff?

@harvey @FireworksAI_HQ What is your post training approach? Did you ensure the model remains conditioned not to confabulate? Did you run an experiment asking these models to explain a made up law and see where they differ? We found fine tuning removes the “I don’t know” post training

We partnered with @FireworksAI_HQ to train open-source models for legal. Here's what we found: 1) Hybrid legal agents can beat frontier models on quality and cost by routing selectively to a frontier advisor. We tested a hybrid setup where GLM 5.1 served as the primary worker, routing tasks to Opus 4.7 as an advisor when needed. GLM invoked Opus sparingly, just 0.83 times per task on average. The hybrid setup beat Opus on both quality and cost: 18% all-pass vs 14%, at $368 vs $954 across the same 100 tasks. 2) Post-training can push open models to frontier-level legal performance. On a 100-task slice of our Legal Agent Benchmark (LAB), SFT moved Kimi 2.6's all-pass rate from 11% to 15%, beating Opus' 14%. But the cost gap was even more striking: $84 vs $954 across the same 100 tasks, or ~11x cheaper. We're excited to continue working with @FireworksAI_HQ on the next generation of open-source legal agents.

Whatever happened to Google Jules?

@AndrewYNg Sounds like QB from McKinsey or Gamma BCG have the better FDE value prop given your independence wish. If only they paid the same as the labs they’d have the same talent levels too. I predict FDE will be far bigger in EU than US because consultants are really common here.

One of the new, buzzy jobs in Silicon Valley is the AI Forward Deployed Engineer (FDE), an engineer who is embedded within a client organization to help customize solutions, such as building and tuning agentic workflows that suit the client’s particular needs. I’ve heard from people who are wondering anew about the FDE career path since OpenAI and Anthropic started building new teams to place FDEs within client organizations. The rise of FDEs for AI workloads is one way AI is creating new jobs (and why the jobpolcalypse narrative of upcoming job market collapse is false -- there will be many AI and non-AI jobs). However, I believe there will be far more AI Engineer jobs than FDEs, as I explain below. The FDE role was pioneered about two decades ago by Palantir, which sent engineers to government locations to work on secure, air-gapped networks. In addition to having good technical skills, FDEs need communication skills and sometimes business skills. For example, they may need to speak with clients to understand their needs, formulate a strategy to prioritize projects, explain complex technology, and respectfully push back if a client asks for something unrealistic. They’re enjoying a resurgence because of the amount of work involved in taking an off-the-shelf LLM and building it into a custom agentic workflow that fits particular business needs. However, I believe the number of AI Engineer jobs will be far larger. A company might accept a few FDEs to be embedded within its organization. But most companies will want far more of their own employees working on their projects. While my organizations do hire FDEs, we hire far more AI Engineers! Also, a common client concern is that it is hard to find vendor-neutral FDEs — they are, after all, there to deeply integrate a particular vendor’s product into a company. In this moment when it’s hard to predict which AI service will be the best one in a year’s time, optionality (the ability to pick whatever vendor turns out to fit best in the future) is very valuable. In contrast, letting FDEs tightly bind a company’s processes significantly reduces optionality. Right now, I see surging demand for AI Engineers who can build software applications using AI software components (like LLM prompting, agentic frameworks, evals, etc.) and effectively use AI coding agents (like Claude Code, Codex, Antigravity CLI, and OpenCode). As the AI Engineer role matures, I expect it to fragment into more specialized roles, like the generic Software Engineer role from decades ago fragmented into frontend, backend, mobile, data engineering, devops, and so on. What will be the future, specialized AI engineering roles? I don’t know. Perhaps there will be AI FDEs, LLMOps Engineers, Evals Engineers, AI Data Engineers, Harness Engineers, and other roles we don’t have names for yet. But for now, I see a lot of AI engineers who are generalists create a lot of value. Skilled AI Engineers are in very high demand! As our field continues to mature over the coming decade, I look forward to new specializations within AI Engineering that create even more job opportunities. [Original text: The Batch newsletter]

@ArtemXTech Lol I'm so happy I ran this before :D github.com/pascalwhoop/co…

Your Claude Code agent traces are the most valuable data right now. But they might be deleted tomorrow. There is a setting in Claude Code called cleanupPeriodDays. By default, it cleans up your conversation history after 30 days. Sessions older than that get deleted automatically. It is recommended to set it to a large value. I set mine to one year and I back it up. What you can do with that history: - Extract insights from your agent sessions. - Teach your agent not to repeat. - Avoid the mistake of repeating yourself. Leverage your data, which you truly own. Docs: code.claude.com/docs/en/settin…

@PSkinnerTech @NessieLabs Wait, is Nessie basically github.com/pascalwhoop/co… but with a paywall? Wild. Like the problem is absolutely worth solving! But idk man, 80$ just so you can access your own chats via MCP/CLI? nuts. Just put it in a git repo

Hot take: obsidian is a terrible form factor for an AI 2nd brain. GBrain is functionally best option, but non-technical users can’t reconcile that they can’t “see” what the AI knows. Very soon, something like @nessielabs will become the top 2nd brain option.

@dlwiest I built McKinseys API gateway and everyone’s using it over there to codex and Claude code. Everyone sits under one pool of usage

Can anyone explain to me why companies don’t just give employees $100 / month Claude Code or Codex plans instead of paying per token? There has to be an explanation, because this keeps happening and doesn’t make sense otherwise

@dlwiest Few reasons. 1) vendors approvals processes are slow. API is 100x faster to do than SAML enterprise app. 2) 80/20 usage pattern. Most barely use it. You pay seats for idle people. 3) super users still need 100x more tokens

@TradexWhisperer I have a list of lab in the loop companies. People selling reactants and sequencers. Most of these went really down after Covid. But IMO going after “the living world” is holy grail. If you can mold that through AI you can change anything

Where's the fun in that. Too obvious. I support difficult names. Micron 30-year cyclical business, I am defending that it is no longer cyclical. Palantir Forward PE ratio of 100, I am still defending that it will take over the Enterprises and Governments. Rocketlab, barely making any money, I gave a price target of $100 last year. Look where they are now. Microsoft is being hated literally by everyone. I defend that their distribution power will be key to Agentic AI. Crowdstrike. Forward PE of 138. I am saying Agentic AI will be a REQUIREMENT to defend against future AI attacks. ServiceNOW, nobody know what the fuck they do. their software is pretty much invisible to the retail yet I support their digital workflow. Difficult to defend their thesis but it's fun.

The Agentic AI era will produce the biggest software winners in the next decade. Which software name wins the Agentic AI era? Mine is in the comments. Drop yours too.

@TradexWhisperer Ah I see. To me Reddit Microsoft and SNOW are as known as Google but then again I’m an enterprise guy for 10+ years. My bets are on picks and shovels in the biological world. CPU/RAM went up cuz learning loops for agents. What’s equiv for programming biology?