Nathan Stevenson

AI Builder: founder @ Forwardlane | 7 Patents | Engineering AI that scales | Stealth startup ~ code intelligence and verified agentic trajectories

[Nathan Stevenson @NathanStvnsn]

@dunkhippo33 @dunkhippo33 Is Hippo Campus open today ? Our team is excited to come and cook this afternoon before Pat arrives.

[Parul Gautam @Parul_Gautam7]

We're getting to the point where AI isn't just helping us write code; it can help launch the business behind that code too.

One of the more interesting examples I've come across is **Lovie**

Instead of jumping between legal websites, paperwork, and incorporation services

Lovie lets founders form a US LLC or Corporation directly from AI coding tools like **Claude Code, Cursor, and Windsurf** through its MCP-first workflow.

If you're already building your product inside your AI coding assistant, the setup process can become part of that same workflow instead of another tab to manage.

For founders, indie hackers, and developers shipping fast, that's a pretty compelling idea.

Flat **$29/month**, company formation in minutes, and it fits naturally into the AI tools many developers are already using.

Feels like another step toward AI becoming a true operating system for builders, not just a coding assistant.

[Sakana AI @SakanaAILabs]

"Feedback-to-Rubrics: Can We Extract Expert Criteria from Inline Comments?" will be presented at the Workshop on Human-AI Co-Creativity: Advances, Opportunities, and Challenges on July 11 at #ICML2026

Paper: https://t.co/MPfNX6l2z1 (Extended full-paper version)

LLMs are increasingly used for writing and review support, but their usefulness depends on context-dependent criteria, such as expert preferences or organization-specific conventions, that are often tacit, undocumented, and difficult to elicit directly. We propose a problem setting for learning reusable natural-language rubrics from accumulated inline comments on artifacts such as human-written or LLM-generated drafts.

[Deedy @deedydas]

“The dirty secret in AI is that everything is a data and an eval problem.

The best models have the best data and best internal benchmarks. The mid ones buy a lot of data, not the best, and hillclimb public benchmarks.

(you need a lot of compute too)”

– Stanford CS Professor

[Nathan Stevenson @NathanStvnsn]

@dan__rosenthal MICROSITE

[Nathan Stevenson @NathanStvnsn]

@realAkashAnand Research

[Nathan Stevenson @NathanStvnsn]

@steipete @aiDotEngineer Where did you download the sessions from?

[Nathan Stevenson @NathanStvnsn]

@jun_song @steipete @Alibaba_Qwen @Alibaba_Qwen is this really the message you want to deliver to the world?

[Nathan Stevenson @NathanStvnsn]

Instead, he has to deal with your tiny small person ego trying to present what to the world? It’s a shame to @Alibaba_Qwen that you are supposedly an ambassador.
Many and MOST Chinese and Korean AI labs, researchers are humble, are extremely intelligent and talented — and they are KIND and SUPPORTIVE to the industry.

[Nathan Stevenson @NathanStvnsn]

@mikefutia Reddit

[Nathan Stevenson @NathanStvnsn]

@VittoStack Interested!

[Nathan Stevenson @NathanStvnsn]

@NickAbraham12 Leads

[Nathan Stevenson @NathanStvnsn]

@elonmusk Yeahaa!!

[Nathan Stevenson @NathanStvnsn]

Started building this a year ago out of frustration with scattered data and desire for real provenance - reality is large code bases (150+ repos), office documents (word, ppt, xlsx) , workplace tools (slack, Jira, g suite, asana, Salesforce, Hubspot) are different animals requiring different approaches. Snowflake does not cut it, in fact OLAP is a problem because in a real business there are transactions and massive amounts of them. The solution lives in ontologies, real-time context graphs around user and the business / work lens - then agents can live and work across the enterprise substrate and provide services and workflows to user and business. Any interface works. Ontologies help determine what is meaningful and what is noise. This is our approach.

[John Suh @john_ssuh]

Increasingly, I believe companies may need to be rebuilt from the ground up, where you have a single timeline of all observability + product metrics + file changes laid out in a retrievable system, like Datadog + Posthog + Google Drive + Slack (really unified filesystem of Claude Code chats + Codex chats). This might be the new data foundation for any and all companies to maximize AI. Needs to be rebuilt because keeping track of diffs on existing system basically impossible to produce longitudinal information on decisions and rollbacks, something coding agent storage companies are actively trying to figure out, but this should extend to businesses as a whole.

Highly skeptical existing businesses will adopt this though because it means overhauling everything about their instrumentation and business data, but I think businesses built on this foundation probably can execute 100x better and faster

[Nathan Stevenson @NathanStvnsn]

This is part of the solution to enterprise substrates - MCP is latency, tool calls are bottlenecks to context, but faster retrieval is only one part of the stack.

[Emil @emilsnotes]

Today we're launching Kapa for Agents: all your product knowledge, in one tool call.

As models get smarter ... and pricier ... (looking at you Claude Fable), context is becoming the bottleneck. Here's why:

Say you've built an agent in your SaaS app with some tools. Then a user asks "How do I enable SSO?" but you haven't given the agent a tool to fix that. It doesn't matter how smart your agent is, it hits a dead end.

Instead, with Kapa, you can add easily add a single knowledge search tool so your agent can read your docs, code, tickets.

Real-world agents use this in +40% of interactions to improve planning and avoid dead ends.

TL;DR: we spent 3 years building the best agentic retrieval platform that:

→ Finds the right source ~2x more often than web search or a DIY RAG pipeline

→ Tells you what your agent couldn't answer, and exactly how to close the gap

→ Connects 30+ sources in one click, synced in real time so knowledge doesn't go stale

→ Works with any agent: product copilot, support agent, or Claude Code

Teams like @tweetsbyport, @airbytehq, @circleci and @matillion are already using Kapa's Retrieval MCP and API to build in-product copilots, support agents, RFP tools and coding assistants.

... and today, its available to everyone at kapa [.] ai / agents.

[Nathan Stevenson @NathanStvnsn]

Nice ideas for loops

[Nick Dobos @NickADobos]

Loops so hot right now
Here are the loops I’m experimenting with:

Planners
- morning brief, orient and context switch back into the flow. This pulls from outside data sources and also checks past chat logs to figure out our current progress.
- daily planning
- what did you get done this week? Friday end of week check in & review
- Sunday/monday planning, 3 lanes, personal, side biz & projects, day job. Goal is top task prioritized and clear.
- Monthly planning. Personal & side. business check in. Go over finances, etc.

Once weekly codebase / promptbase sweeps
- code maid & janitor
- code & prompt evolver & compounder
- memory and docs cleanup consolidate dream sequence sweep

Watchers
- sentry -> auto crash fix. Runs once a week and priorities top crashers
- analytics alert -> auto fix. Fires when analytics trip. Usually when something breaks.
- triager, scan bug reports and customer support requests from linear. Either fix or defer. When a fix is merged/shipped email the person who reported it.

Stock Day trader
- once a day check portfolio, do research, check news, copy politicians, rebalance if needed

What loops are you using?