Erasmus Elsner 🦊

tweeting about early stage tech, the founder journey and venture capital, EIR @MindsDB

[Erasmus Elsner 🦊 @erasmuselsner]

Silicon Valley Dissatisfaction Cycle:
🧑🏼‍🚀->🤵‍♂️founders want to be VCs
🤵‍♂️->👨🏻‍🎤VCs want to be influencers
👨🏻‍🎤->🧑🏼‍🚀influencers want to be founders

[John Felix @johnfelix123]

In my early 20s I was lucky enough to have a call with Chris Dixon and asked him what advice he had for me. He said something similar - that most LPs are too rigid and need their GPs to fit into a box. And instead to focus on finding special investors, build deep trust with them, and let them do their thing, no matter how weird and far outside the box it might seem.

[Yash Bhardwaj @ybhrdwj]

I turned sam altman's texts to mira murati into 2011 style emo teenage heathrob anthem🫶

[Endowment Eddie @endowment_eddie]

Is your teen texting about complex LP/GP dynamics?

Cc: @PEoperator @TheIcahnist

[Riley Walz @rtwlz]

About to annoy so many people

[Erasmus Elsner 🦊 @erasmuselsner]

The purpose of venture classic is to let founders run experiments against hypotheses that are knowable and testable, but uncertain.

[Kyle Harrison @kwharrison13]

https://t.co/LvmU99CyIO

[Om Patel @om_patel5]

stop spending money on Claude Code.

Chipotle's support bot is free:

[Trace Cohen @Trace_Cohen]

Vertical AI? Of course that’s your contention. Of course it is.

You just finished watching a Sequoia deck breakdown on “AI for Law” and now you think you’re going to disrupt Deloitte with a few fine-tuned prompts and a clever GPT wrapper.

You’ll believe that right up until next month when you crack open Sutton & Barto and start throwing around “policy gradients” like you just invented reinforcement learning, quoting enterprise API limits like they’re state secrets.

Then you’ll finally talk to a real GC or a hospital CIO and realize your “vertical AI” isn’t an LLM with a moat — it’s a brittle workflow taped together by LangChain and Zapier. You’ll start quoting HIPAA clauses, SOC2 reports, and latency budgets like gospel while your RAG pipeline silently dies because the data you need lives in some 2004 SharePoint server.

After that, you’ll get real ambitious, quoting Bessemer’s State of the Cloud and pretending you actually understand CAC:LTV ratios when your model inference costs are eating your gross margin alive. You’ll cite “fine-tuning efficiency” and “domain specificity” while your token burn rate climbs faster than your MRR.

“Well, as a matter of fact, I won’t, because Vertical AI is the next SaaS wave. Every industry gets its own copilot. The TAM is enormous—”

Right. The TAM. The slide every founder prints before they have a single paying customer. I’ve seen that movie. We called it “SaaS 2010.” It ended with 50 identical dashboards selling into the same ten budget owners.

That’s not a moat; that’s a mosh pit.

Is that your thing now? You read a16z’s “Who Owns the Vertical?” post and suddenly you’re a prophet of industry transformation?

You start throwing around words like “semantic layer,” “closed-loop feedback,” and “multi-agent orchestration” to impress LPs who haven’t logged into Notion since 2019?

One, don’t do that.
Two, you dropped a $500K pre-seed check into a wrapper app that could’ve been invalidated by a weekend of customer discovery.

“Well, at least I’m betting on founders who ship.”

Yeah, maybe. But I’m betting on founders who understand — who’ve lived the problem, who know the regulations, the sales cycles, the data formats, and the people who actually write the checks. The ones who know that “vertical” doesn’t mean “small TAM,” it means defensible domain expertise.

First principles isn’t about duct-taping a chatbot onto a workflow. It’s about asking whether this industry’s data, process, and trust layers can even support automation yet — and if so, what new infrastructure has to exist to make it real.

But hey, if you’ve got an issue with that, we can always take it up with E.

[Kalshi Finance @Kalshi_Finance]

I'm hearing that a Series C AI infrastructure startup is telling employees they have "18 months of runway" while quietly shopping for acquihires and the founder is already interviewing at FAANG companies

Source close to the situation says they're doing "voluntary unpaid sabbaticals" and calling it a "company-wide mental health initiative" - meanwhile the entire go-to-market team got managed out last week

The grift is finally catching up to these people who spent three years building glorified ChatGPT wrappers while pretending they were the next Google. Time to learn actual engineering skills instead of prompt engineering

[Antti Karjalainen @aikarjal]

Someone needs to build a company around Customer Context Graph.

Collect all the threads – emails, meeting transcripts, slack messages, contracts, deliverables, detail, info, and config – from your customers into context that can be explored and queried by agents.

This info is scattered between CRMs, ticketing systems, note takers, product, landing pages – it's inherently cross platform information. You need a new solution. Kind of how Segment did it trad SaaS apps.

With this context, you can fire up Claude Cowork or similar for ad-hoc work or build extremely powerful agent automation flows. Expose the context as skills, MCP, and file system.

Even better if you build it as open-source with a hosted option so people can take it on-prem as needed. Create a connector ecosystem around it. This will power every single next-gen AI-native full-stack business.

Sort of like the context graph (@ashugarg @JayaGup10 ) that has been discussed recently but I'm thinking something very concrete: "Get me all the context about this particular customer."

A customer-level, cross-system context substrate that agents can explore and act on

[Megan Anne Agathon @agathomai]

an article I wrote on how consuming ai generated content changes our perception of reality, making outliers imperceptible, and accelerating our culture towards the mediocre. enjoy ❤️

[Erasmus Elsner 🦊 @erasmuselsner]

"portfolio construction"

[ash @asharoraa]

Bet-the-fund moments are Silicon Valley success stories:

- @A16z bet 17% of their fund in @Skype
- @Sequoia bet 15% of their fund in @WhatsApp
- @ThriveCapital bet 20% of their fund in @Stripe
- @GVteam bet 22% of their fund in @Uber

They all succeeded, hence the funds became/ stayed tier 1

@FoundersFund just bet 17% of their fund into @Cognition. They bet 30% of the last growth fund into @anduriltech

🔥

[Dan Gray @credistick]

VCs tried to scapegoat @linamkhan for the post-2022 M&A slowdown.

Now they’re trying to blame the FTC for sketchy acqui-hires.

The problem, in both cases, is just BAD INVESTMENTS. Too many overcapitalised companies with no real moats.

There are receipts.

[Erasmus Elsner 🦊 @erasmuselsner]

don't be the problem, be the solution that creates the problem

[Terrence Rohan @tmrohan]

The founder of a top YC company in the current batch on Venture Capital:

"People used to climb Everest and they needed oxygen. Today, people climb it without oxygen. I want to summit Everest and use as little oxygen (VC) as possible."

Vibe shift.

[Ed Sim @edsim]

Massive barbelling at Inception rounds these days because of AI:

Either founders want to:

Raise <$2M or as little as possible 👏🏼

or

Raise >$10M to go big as cost of entry in certain markets requires big 💰 for talent, training models, etc.

I'm seeing less of the classic rounds of $3-4M

Huge implications for pre-seed, seed firms as this continues - which lane does one stay in?

What does lots of small $1m checks mean for portfolio construction?

Do I just pass on that generational founder going big?

All up for grabs!

[Erasmus Elsner 🦊 @erasmuselsner]

all businesses are loosely functioning disasters, some just happen to make money

[Balaji @balajis]

It might turn out that the GPT wrapper has more of a moat than GPT.

[Percy Liang @percyliang]

While we celebrate @deepseek_ai 's release of open-weight models that we can all play with at home, just a friendly reminder that they are not *open-source*; there’s no training / data processing code, and hardly any information about the data.

[Justine Moore @venturetwins]

The Stanford experience

[Lucas Beyer (bl16) @giffmana]

I've been trying out Cursor with o1 for a few weeks now, and it's been giving me proper "holy shit, this changes things a bit" vibes.

The most impressive to me is not the "generate code for XYZ" you see everywhere. That's nice, but I can also do that myself just fine, so it's only saving me a few minutes.

What really impresses me is when you index the whole codebase, and then use Cursor's "codebase chat" feature. When you're working with a codebase you're not intimately familiar with (which is most of the time for most developers in mid+big companies), you can literally ask it any question/problem about the codebase, and it answers it. It's like having the codebase author(s) at your disposal, you can ask them all dumb questions, and they answer immediately, without judging you and without you wasting their time.

This is insane!! So good. First step-change in the SWE part of my job since a long time.