About
https://dosu.dev
Articles by Devin
-
How to Build Agent Memory: Search & Retrieval
How to Build Agent Memory: Search & Retrieval
This post is based on our Agent Memory: How Does It Find Anything? webinar. Watch the recording or read the full…
1 Comment -
How to Build Agent Memory: Knowledge Capture
How to Build Agent Memory: Knowledge Capture
This post is based on our Agent Memory: What Information Is Worth Remembering? webinar. Watch the recording or read the…
2 Comments -
The Building Blocks of Agent Memory
The Building Blocks of Agent Memory
There's a lot of noise right now about agent memory, agent knowledge, and "company brains." Everyone agrees agents…
6 Comments
Activity
3K followers
-
Devin Stein shared this🚨Calling all Super Smash Bros fans 🚨 #SFTechWeek is next week and we’re hosting a Super Smash Bros night before the panels start. Join us at Dosu's office Monday 10/5 for a fun night of both Super Smash Bros. and agents. No skill required. Items on, stages random (or not 😉). Register: https://lnkd.in/eUsi9y3C
-
Devin Stein reposted thisDevin Stein reposted thisMost AI memory benchmarks test whether an agent can recall a specific piece of information. But that’s more like pure search and not really how coding agents work or what they even need to be good at. The best coding agents need to inherit messy histories, encounter related problems across thousands of interactions, and learn from what happened before. And for these reasons, Michael Mangus put VibeMemBench alone in his S-tier. Here’s why he thinks it points toward a much more realistic way to measure agent memory. Full transcript in comments👇️
-
Devin Stein shared thisWe ran the new models through our coding agent token usage benchmark. OpenAI's GPT-6.1 Sol vs Anthropic's Claude Sonnet 5.5 and Claude Opus 5.5. For every one of them, the biggest line item was context. GPT-6.1 Sol put 58% of its cost into context and only 3% into planning. It finished all 5 tasks for $0.45. Sonnet 5.5 and Opus 5.5 spent 40% and 38% on context and about a quarter on planning. Opus cost $1.58 (~3.5x Sol). Sonnet cost $0.72 (~1.6x Sol). Measured with Decant https://lnkd.in/eWpxHv4E
-
Devin Stein shared thisFor those that are observant, you'll notice that I'm one table down getting demolished by an LanceDB intern. That same intern played blindfolded and won later in the night. This was at Dosu's first Chess Night and it was so much fun that it has revived my lichess app addiction 😬 s/o to George Xu and Jono Sison for putting it together!Devin Stein shared thisLast week, I had a great time competing in Dosu's Chess Night. There were some really highly rated players, including FIDE Masters, and I am happy to have won a mechanical keyboard from the event! Big thank you to Dosu and George Xu for inviting me. The games devolved into a few of us playing bughouse matches. Here is a pic of me learning from legendary bughouse players Misha Lubich and João Sousa-Pinto. I always end up meeting the friendliest and smartest engineers at these chess events. Hope to join the next one!
-
Devin Stein reposted thisDevin Stein reposted thisLast week, I had a great time competing in Dosu's Chess Night. There were some really highly rated players, including FIDE Masters, and I am happy to have won a mechanical keyboard from the event! Big thank you to Dosu and George Xu for inviting me. The games devolved into a few of us playing bughouse matches. Here is a pic of me learning from legendary bughouse players Misha Lubich and João Sousa-Pinto. I always end up meeting the friendliest and smartest engineers at these chess events. Hope to join the next one!
-
Devin Stein reposted thisDevin Stein reposted thisWhat will it take for enterprises to run software engineering like AI-native companies? Today, we are announcing Autoheal's $𝟳.𝟵𝗠 𝘀𝗲𝗲𝗱 𝗿𝗼𝘂𝗻𝗱 to help make it possible 🚀 This funding will help us expand our go-to-market team, accelerate enterprise rollouts, and strengthen the platform reliability and customer support that enterprise adoption demands. Our thesis is simple: enterprises need 𝗮𝘂𝘁𝗼𝗻𝗼𝗺𝗼𝘂𝘀 𝘀𝗲𝗹𝗳-𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁 and 𝗳𝗶𝗻𝗲-𝗴𝗿𝗮𝗶𝗻𝗲𝗱 𝗴𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 working together to trust agents with more engineering work. That’s why we built Autoheal: a 𝘀𝗲𝗹𝗳-𝗶𝗺𝗽𝗿𝗼𝘃𝗶𝗻𝗴 𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗳𝗮𝗰𝘁𝗼𝗿𝘆 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲. We provide prebuilt, cloud-based worker agents for incident response, vulnerability remediation, and AI coding cost efficiency. We also provide the sovereign infrastructure enterprises need to build and continuously improve agents across the post-coding SDLC, all within their security boundary. Here’s how the self-improvement loop works: ↳ 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗲. Analyze agent traces against signals from code reviews, CI failures, and production incidents. ↳ 𝗜𝗺𝗽𝗿𝗼𝘃𝗲. Turn those signals into proposed changes to agent skills, prompts, tools, or model selections, then check historical benchmarks for regressions. ↳ 𝗚𝗼𝘃𝗲𝗿𝗻. Engineers control access, govern actions, and approve changes to agent behavior. Those changes are version-controlled and auditable. The goal: free engineering teams from 𝗽𝗼𝘀𝘁-𝗰𝗼𝗱𝗶𝗻𝗴 𝘁𝗼𝗶𝗹 so they can build more revenue-generating software. We couldn't do this alone. Thank you to Harpinder Singh at Innovation Endeavors for leading the round, alongside Anupam Rastogi at Emergent Ventures, Kim 💚 Moore at U&I Ventures, Amir Shevat & Raphael Danilo at Darkmode Ventures, Dheeraj Pandey & Amit Kumar at Param Hansa Values LLC, and Quang Hoang at Batch Ventures (CTO Fund). We are also incredibly grateful for our key angel investors: Shawn K., Sumeet Arora, Anshu Sharma, Savin Goyal, and Srikant Gokulnatha. Thank you for backing our vision so early. A heartfelt thank you to our special customers: Sameer Jain at Nomura Bank, Krish Shetty at AvidXchange, and Vijay Pendyala at Empiric Earth for helping us turn our shared conviction into a product. And to Team Autoheal: thank you for earning our customers’ trust and building an incredibly deep product in such a short time. For leaders already expanding agentic engineering beyond coding, what gives you the confidence to trust agents with greater autonomy? I would love to hear what’s working and what’s still missing. Our blog is in the first comment below.
-
Devin Stein reposted thisDevin Stein reposted thisIt's Dosu Drop Day! 🎒 For our back-to-school edition, Dosu does the studying. When a coding session ends, Dosu reads it and writes down what your team would otherwise rediscover. Here's what's new! 📝 When your session ends, Dosu starts studying: The Dosu CLI now installs a session-end hook for Claude Code, Codex, and Cursor. When a session ends, Dosu studies it in the background and saves useful decisions, gotchas, and how-things-work notes to your Library, ready for your team's next task. Run `dosu upgrade`, then `dosu setup`. Setup installs the hooks, refreshes your agent skills, and offers to study your existing session history. 🧾 See what your agent read: Each time your coding agent calls `read_knowledge`, Dosu shows a Knowledge read card with the Sources and notes it returned. The card appears in Cursor, Codex, Claude Desktop, and ChatGPT, and terminal clients get the same list as text. If you've connected the Dosu MCP, the read card works for you today. 💳 Extra credit: Before a full Knowledge review, Dosu checks whether a merged pull request could affect your Library. Changes limited to CI config, lockfiles, or generated files use no credits, and a PR that needs no update costs 2 credits instead of a full review. We also shipped explanations for messages Dosu left unanswered, admin access for individual Agents, Library and Agent management in the Dosu CLI, and Analytics across All Libraries.
-
Devin Stein shared thisPeople often ask what is the team like at Dosu and I like to say "kind nerds with cool hobbies" because it's true. Michael Mangus is not only a champion debater, has PhD in computational social science, is exceptional engineer and product thinker, but he is also an incredibly talented film photographer. Most people pay have to pay for his work, but if you come to chess night this Thursday, you have the opportunity to win this print for free :) https://lnkd.in/eFatKTSfDevin Stein shared thisi made a cyanotype print from an 8x10 film negative for the winner of tomorrow's Dosu chess event. if you recognize Qb8+! Nxb8 Rd8# you should sign up: https://luma.com/dosu-7bzh (I would like to personally apologize for the board in the luma banner being rotated wrong)
-
Devin Stein reposted thisDevin Stein reposted thisThe team keeps working after the agent’s session ends. A pull request merges. A support conversation gets resolved. Someone closes a ticket. Devin Stein explains why agent memory needs to react to those external events to stay current across an organization. Updating memory only when the agent learns something leaves a lot of the team’s work out of the picture. A clip from episode 5 of our agent memory series. Recording and transcript in the comments.
Devin Stein replied to a comment
10h
Hakan D. Great catch. LLM math 🙃
-
Devin Stein liked thisDevin Stein liked thisMost AI memory benchmarks test whether an agent can recall a specific piece of information. But that’s more like pure search and not really how coding agents work or what they even need to be good at. The best coding agents need to inherit messy histories, encounter related problems across thousands of interactions, and learn from what happened before. And for these reasons, Michael Mangus put VibeMemBench alone in his S-tier. Here’s why he thinks it points toward a much more realistic way to measure agent memory. Full transcript in comments👇️
-
Devin Stein reacted on thisWe hosted our first in-person event at the Dosu offices last Thursday and had a ton of fun playing all different kinds of Chess! There were lots of competitive play mixed with casual games and everyone who showed up brought great energy and enthusiam 🔥 We had players rated 2300, 2100, 2000, 1900, and plenty more across the spectrum. It wasn’t only strong players grinding each other. There were a ton of expert–beginner games where the stronger players were genuinely happy to share how to get better. I went 0–5 playing against experts and I’m grateful for every lesson they gave on how to improve my gameplay. Thank you to everyone who came. The next one is going to be even better. 👊 Special s/o to Nikita for bringing a bunch of extra boards to accomodate the additional sign ups we had! And congrats to Nikita and Jaisuraj Kaleeswaran for a great game in the finals! If you were there and have a favorite board or moment, drop it below (or tag a partner you played). And if you missed it, keep an eye out as we’ll be hosting more soon!
-
Devin Stein reacted on thisDevin Stein reacted on thisVery excited to share that I'm joining Harvey! 🎉 Over the past couple years, I’ve seen how AI is completely transforming the software industry. It's clear legal is going through the same shift right now, and Harvey is leading it. The work is high-stakes and the quality bar is sky high, which makes it one of the most interesting engineering problems out there. Thrilled to work with Nico Belmonte, Siva G., Sridhar Reddy, Steve Mardenfeld and the team to help build what's next!
-
Devin Stein reacted on thisDevin Stein reacted on thisBack when I was in high school, I was not in chess club. Last Thursday, I was at Dosu's chess night and lost three games in a row 🫠 There were a lot of highly rated players and fast games. Next Monday (Oct 5), we're hosting another game night at the Dosu office: Super Smash Bros., Scrabble, chess, Codenames, and more. Sign up here if you'd like to join: https://lnkd.in/gcSPBJ6x Follow our events page for future Dosu community events: https://lnkd.in/gU-c2Fcg Me about to blunder 👇
Experience & Education
Volunteer Experience
Projects
-
Music for Me
-
To simplify my music discovery, I developed a Facebook Messenger bot to scrape Bandcamp for trending albums and deliver me a new album every morning:
• Use Node.js to develop the bot and MongoDB backend to store songs I like
• Heroku for hosting and to schedule the app to run every morning
• Used Cheerio.js to scrape the website for data
• Facebook Graph API
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content