Lance Martin
@RLanceMartin
Followers
15K
Following
6K
Media
417
Statuses
1K
langchain. past: robots 🚘 🤖, phd @stanford 🧪
San Francisco, CA
Joined May 2009
Common “context engineering” patterns. Loved @dbreunig posts on this. I also wrote up some thoughts:.
As your context bloats, you hit different failure modes. These failures hit agents hardest because they operate in exactly the scenarios where contexts balloon: gathering information, making sequential tool calls, engaging in multi-turn reasoning, & accumulating histories.
3
7
28
Gemini2.5 video understanding + text-to-speech are very good. Simple multi-modal researcher I put together using native web search + YouTube video understanding tools w/ text-to-speech. Researches topic + analyzes videos, produces report + custom podcast.
2
4
26
Building Async ("Ambient") Agents. Happy to share new, free course on building "ambient" agents! This is one of the most interesting agent UX patterns (e.g., Devin, Codex), allowing the agent to do work "in the background" and interact with the user via human-in-the-loop for
5
62
364
Some useful references --.1/ @karpathy on LLMs as OS.2/ @walden_yan on context engineering.3/ @barry_zyj + team multi-agent.4/ @AymericRoucher + team on deep research.5/ @bcherny on.
1
6
34
I wrote about some popular patterns for managing context ("context engineering") w/ AI agents: .
5
56
400
a few thoughts on the current state of agents based on what I saw at @aiDotEngineer: . rise of "ambient" agents. the bitter lesson & agent UX. RL for non-verifiable tasks. the case for MCP. early days for agent memory .
5
19
115
@kevinhou22 on Windsurf:.> Current dev workflow centric .> Highly opinionated UI / IDE.> Allows for granular data capture .> Lets them train models. @mntruell w/ @benthompson pod similar point; long “messy middle” of devs + AI working together preserves need for IDE.
0
0
6
Most interesting AI product question I took from @aiDotEngineer is Claude Code vs IDEs (Cursor/Windsurf). @bcherny on Claude Code: .> Bitter lesson centric .> General models win.> General things around model win.> Unopionionated / no UI.> Work w fast changing UX / models
2
1
39
Some notes from @aiDotEngineer day 1 -. @simonw on state of AI.> Visual eval for LLMs: asked each LLM to generate code for an SVG image of a pelican riding a bicycle. Ran this across ~30 model releases over the past 6 months. Created a script to select random image pairs, GPT4.1.
5
19
114
Agent can be hooked into Gmail by swapping out the tools used. Components are also general and can be used w/ various tools / MCP servers.
0
0
2
Memory -- Add memory, so the agent learned email response preferences from human feedback. Notebook:.
1
0
5
Human-in-the-loop -- Add human in the loop for approval / editing of specific tool calls. Notebook:.
1
0
2
Agent evals -- Unit tests (Pytest) for triage decision + tools calls (test structured outputs using heuristic eval) and LLM-as-judge to eval email responses. Notebook:.Slides:.
2
0
5
Building agents -- Combing workflow (router) with agent that can call email tools. Notebook:.Slides:.
1
0
2
Fundamentals -- Tool calling, agents v workflows (h/t @barry_zyj, @ErikSchluntz), persistence/checkpointing. Notebook:.Slides:
1
1
8
Agents from scratch. This repo covers the basics of building agents:. Fundamentals . Build an agent. Agent eval. Agent w/ human-in-the-loop. Agent w/ long-term memory. Builds to a deployable agent to run your email. Code (all open source):.
9
109
706