Ramy Adeeb

Big, proprietary Foundational Models - YES! Tuned & customized models for enterprises - YES! Inference & AI Infra providers - YES! Big 3 CSPs - YES! On-device inference - YES! I guess it is human nature to think zero-sum (SAD!), but sure feels like there's demand for all of it.

Q: Why do we need data centers in space ? A: Because terrestrial hype cycles were getting too crowded

@rabois @rex_woodbury it's also objectively a better place to live -- something many founders care about

Provocative post by @rex_woodbury on why NYC is leading geo for building successful AI application startups: digitalnative.tech/p/new-york-is-…

Somehow LLMs are starting to reflect the personalities of their founders: Claude — thoughtful, slightly anxious honor student ChatGPT — extremely eager sycophant Groq — unhinged

@ycombinator @insforge Thank you Y Combinator 🫡 Check us out here guys: insforge.dev Our repo: github.com/InsForge/InsFo…

Good articulation of an impatience phenomenon that has infected AI folks.

Two years ago @cline shipped one of the first real coding agents. Today they open-sourced the entire rebuilt engine. Same engine that tops Terminal-Bench 2...Same plugins, sub-agents, cron jobs, and multi-provider routing that powers their own CLI + VS Code extension.

We found a zero-authorization vulnerability in an a16z-backed DoD startup that exposed the data of active U.S. military personnel. We tried to report it. They ignored us for 150 days. Here is how our open-source AI agent found the ultimate OPSEC nightmare 🧵👇

Worrying that your startup will be eaten by the model companies is like worrying that your life will be constrained after you become a movie star. You're far more likely simply to fail.

Here's YC's official advice about being truthful and precise about what is pilot, bookings, revenue and recurring revenue. Founders, particularly first time founders, need to sear this into their brains. Don't mistake one tier for another. Be precise, and always be truthful.

So much Rust on GitHub now, feels like every other project is written in Rust. Did AI do this? Was the unlock for getting engineers to write Rust making it so they don't have to write Rust?

deepseek v4 is now the cheapest sota model available at 1/20th the cost of opus 4.7. for perspective, if uber used deepseek instead of claude their 2026 ai budget would have lasted 7 years instead of only 4 months.

Which emerging VCs have the strongest early-stage picking alpha? Standard emerging manager evaluation still leans heavily on qualitative signals – GP background, thesis articulation, founder references. All useful, but by the time TVPI and DPI tell you something meaningful, you're usually already in or already too late. So I experimented with a quantitative framework to answer a core LP allocator question: which small, early-stage fund managers consistently back seed-stage companies that go on to raise exceptional Series A rounds – before those outcomes are visible to the broader market? I started with @harmonic_ai Scout (my fav research tool!) and checked every company globally that raised a first pre-seed or seed round between 2022–2026 (Post-ZIRP). The funnel looks like this: 1/ 55,491 companies raised a pre-seed or seed round – the full opportunity set 2/ 4,368 (7.9%) went on to raise a Series A – the base rate, roughly 1 in 13 3/ 764 (1.4%) qualified as Tier 1 Breakouts – above-median Series A for their vintage year, with at least one top-tier institutional VC (from a defined set of 38 firms: @a16z, @sequoia, @lightspeedvp, @IndexVentures, and peers) For each of those 1,604 companies, I traced back to every investor who backed them at pre-seed or seed — before the outcome was visible. 4,176 unique investors across the breakout set. Then I computed a simple ratio for each: breakout companies backed at seed divided by total seed investments in the period. I'm calling this the "Tier 1 Concentration Rate". After filtering out mega-platforms, accelerators, CVCs, and angels and requiring a minimum of 10 seed deals – 20 emerging managers (sub-$250M AUM) surfaced with notably high concentration rates. A few things stood out: 1/ Several micro-funds under $100M were placing 25–35% of their seed bets into companies that later raised from @Sequoia, @a16z, @lightspeedvp – consistently, not as one-off flukes. 2/ Participant concentration and lead concentration are different signals. Participant = network and access. Lead = independent conviction before consensus forms. For LP diligence, these deserve to be evaluated separately. 3/ The data has real limitations: ~12% of breakout companies had no named seed investor in the database, we can't cleanly separate Fund I from Fund III for a given manager, and small sample sizes mean some high concentration rates likely reflect luck rather than repeatable skill. But the core idea holds. "Tier 1 Concentration Rate" is an early, measurable signal of picking ability – observable years before fund-level metrics tell you anything. For LP allocators evaluating Fund I–III managers, that timing gap is the whole problem. This is one attempt to close it. What’s your take on this experiment?

We raised $6M to kill the book-a-demo button with self-improving agents, led by @FoundationCap. One customer had Hobbes close a deal end-to-end within an hour of going live. Another has 2x'd revenue since going live. This is the new default for growth in the agent era. Hiring across GTM and engineering in SF. Come build with us.

Talked with Tim Glaser today, cofounder at @posthog! We chatted about early growth, product strategy, open source, and building for developers. Here were my two most important takeaways: 1. The best data you can get is on your existing users. Figure out why they love you or why they churned. Then fix it. Now, your best leads are users who have already signed up once. Reach back out yourself and onboard them personally. This method works 10x better than cold outreach. 2. GTM channels: go deep in one for a month, measure everything. Evaluation and ROI are critical. If it works, double down. Otherwise, stop. Always test new ones because every channel saturates. Got a ton of sharp insights. So thankful for the chat! P.S. We might be cooking up something with InsForge x PostHog 🤫 @insforge

If you are counting pipeline as ARR, you don’t have ARR.

Introducing the new Strix Platform: continuous pentesting for modern apps. Strix is an open-source framework for autonomous pentesting across apps, APIs, and repositories - helping teams find and validate vulnerabilities, generate fixes, and secure software faster. Since our launch, we’ve had: - 80,000+ users worldwide - 15B+ LLM tokens processed daily - 78,000+ vulnerabilities reported - multiple CVEs assigned - deployed by enterprise security teams worldwide Today, we’re launching the Strix Platform for teams that want to run Strix continuously. With Strix Platform, teams can: - pentest their full stack continuously - block vulnerable PRs from merge - validate findings with proof-of-exploit - get merge-ready fixes - retest automatically - track security posture over time Security shouldn’t be your bottleneck. Strix helps you ship faster and deploy with confidence. Try it now 🔗↓ strix.ai/blog/introduci…

A midlife crisis was about buying cars. then instagram turned it into watches Now it's spending $5K vibe coding an app that no one will ever use