7) Whoever solves agent memory will own the next platform shift. It won’t be the smartest model that wins. It’ll be the one that remembers you best.

6) The future of agents depends on memory that is:. A) Selective.B) Inferential.C) Cross-modal.D) Emotionally aware.E) Continuously evolving. Otherwise, we’re just building better autocomplete.

5) We need to train agent memory like we train dogs:. A) Not with syntax, but with signals. B) Repetition, reward, correction. C) Right now, we’re asking agents to be cofounders/employees/etc. when they haven’t even passed obedience school.

4) Cross-contextual inference is non-existent: Agents don’t connect the dots across time, modality, or nuance. You can’t say:. “Hey, remember what I said about Project X and apply it to this hiring plan.”. It won’t. Context isn’t just local, it’s composite.

3) It's missing a sense of "what matters": Agents don’t know what to remember or forget. There’s no memory salience, no prioritization, no hierarchy.

2) It's missing human inference: Ask your agent to recall what you meant last Tuesday in that meeting? It can’t. Why? Because it remembers text, not meaning. Humans infer. Agents regurgitate.

1) Agent memory today is like duct-taped code:. A) It's brittle, hardcoded, and deterministic. B) We don't need "if/then" logic. C) We need inference, intuition, and intent.

Agent Memory is Broken and Here's Why It Matters 🧵:.

7) What market do you think is most ripe for a vertical code-gen tool?.

6) TLDR:. The future isn’t just “AI that builds apps.”. It’s “AI that understands your business.". Vertical-first AI dev tools will win by encoding real-world expertise, not just syntax.

5) Vertical tools are a better GTM wedge. “The Retool for Healthcare Compliance.”.“The AI-native builder for Insurance Risk Teams.”.You get faster land-and-expand, and a stronger moat.

4) Better AI agents come from vertical knowledge. Agents that help nurses, risk analysts, or claims processors need domain expertise. Vertical-first platforms can train better scaffolds, better copilots, and better outcomes.

3) Compliance is a feature, not an afterthought. You can’t fake HIPAA, SOX, or FINRA compliance. Horizontal tools can’t easily adapt to these. A vertical tool can bake this in from day one, which builds trust and accelerates sales. Trust = acceleration.

2) Vertical context = better product-market fit. A Vertical-First Builder Knows:. A. The APIs you’ll use (Plaid, Epic, Snowflake).B. The logic you need (risk scoring, prior auth, KYC escalation).C. The UX patterns that work.

1 Continued ) Horizontal tools like Retool, Vercel AI SDK, and Framer AI are impressive! But they can’t compete with a tool that knows what “KYC flag escalation” or “prior auth routing” means.

1) Horizontal codegen is plateauing. A. Anyone can spin up CRUD apps with GPT + Tailwind now. B. The next layer of value isn’t speed, it’s understanding domain logic. C. A healthcare flow ≠ a fintech dashboard.

🧵 Hot take: The next generation of AI codegen + app dev tools will be vertical-first.

Abominable… at this time of YEAR 😡

Just took the ferry for the first time in NYC. We should absolutely invest more into travel by water, full-stop! IYKYK.

RT @vladtenev: Introducing The Robinhood Chain. The first Ethereum Layer 2 optimized for real-world assets via @arbitrum—from public to p….