About
I'm a strong generalist and call myself a software gardener: someone who guides an often…
Activity
3K followers
-
Beau Gunderson shared thisIn 1995 I was 12 years old, looking through the periodicals at King County Library in downtown Bellevue. WIRED caught my eye: the bold colorful graphics and tech bent designed in a lab to appeal to forward thinkers of all ages. In that first issue I happened to find "The Curse of Xanadu", Gary Isaac Wolf's cover story documenting "the longest-running vaporware story in the history of the computer industry". I later met and befriended Gary through his work starting the Quantified Self movement. He is a fantastic writer and I would recommend anything he's written, but today I am specifically recommending his new book "The Quantified Self: Learning to Observe". Gary believes that self-research is an "important part of the basic human toolkit of reasoning about our problems" and I must agree. Through my own practice of self-research I've managed to solve several of my own problems, most notably the poor quality of my sleep... the single biggest lever for improving the quality of the rest of my life. If you are at all interested in self-research to solve your own problems or improve your health or experience of life this is the best possible starting point. Pre-order link: https://lnkd.in/dUym_v8m
-
Beau Gunderson posted thisi've had three vulnerabilities reported to my most widely-used personal project recently (ip-address, ~4 billion downloads a year, included in the npm and pnpm CLIs and claude code among other things) two were potential SSRFs if someone had used the library as a security boundary to check whether an address was internal or external (in practice i think this is not a common use, but it's certainly worth improving the behavior) one of the reports came in via microsoft due to their partnership with anthropic to scan high-use libraries with mythos (via Alpha-Omega). two of the reports came in through pseudonymous vulnerability hunters. because i did not want to be subject to a slow drip feed of reports from future hunters i decided to run the new claude code security plugin on it (the one released july 22). it found one more vulnerability with a higher CVSS than the three previous, which was unexpected! at least three prior people had used state of the art LLMs, but each likely had a bounded budget. the scan i ran took ~81m tokens and cost $140. probably far above what any individual hunter or anthropic-sponsored team is spending per library! but it makes it very clear that an attacker with a goal in mind can find a gap for a relatively small dollar investment... what have you done to prepare for this new world? what are you going to do?
-
Beau Gunderson reposted thisBeau Gunderson reposted thisThe best way to understand a care modeling platform is to put builders in a room, and give them 8 hours. 👇 That's what Canvas BUILD 2026 was - our first customer event, in San Francisco, and it looked absolutely nothing like a conference. Senior engineers and developers from Canvas sat down with our customers and built together, in real time, on the workflows those teams actually run every single day. It's also where we introduced Studio, which lets teams build directly inside their own Canvas instance - new workflows, note templates, and applications that catch the work that usually has to wait for a developer or slips through the cracks. Eleven years in, our mission is still the same: accelerating everyday medicine. What we heard from customers at BUILD is that Studio changes what they can take on themselves - instead of waiting on a roadmap, they build the workflows and applications their care model needs, and that turns into new services and new products for their patients as they scale. This was the first BUILD, and it won't be the last. The next one is coming soon, you're going to want to be in the room 👀. Come build with us! #HealthTech #HealthcareIT #EMR #CanvasMedical #BUILD2026
-
Beau Gunderson posted thisjust found a devastatingly simple inefficiency in one of our most-used agents (claude-agent-sdk based), making it cost 3x what it should... hit me in the DMs here if you want to know the issue and how to resolve it. why there? because i want to trade tips! looking for other folks doing effective engineering on agentic development and agent workflows to share knowledge with as well as learn from. 💸
-
Beau Gunderson reposted thisBeau Gunderson reposted thisCanvas Studio has been wildly successful. I think there are two reasons why. First is that Canvas end-users and managers tend to have great insight into how to optimize their team's workflows, so giving them an easy agentic tool to make those ideas real is... exactly what the doctor ordered. The second reason Studio has been so successful is that it's built around an iterative development lifecycle: it's not a one-shot prayer for good outcomes, but rather an iterative, stage-gated process that gets users to successful projects (automations, customizations, integrations, etc) within a few minutes, and then collaboratively iterates to a point where all the requirements are met, security reviews pass, and the project can be either handed off to engineering for CI/CD or, under certain conditions, deployed safely to production. To drive the point home, here is a visualization of project maturity and success from a recent cohort of Canvas Studio users. Each dot represents a project (e.g. a tailored population health dashboard, a specialized claim coding automation, a bespoke patient engagement campaign, and so on). Starting a project is so easy that we see many users spin up several at once, and actively flesh out their thinking in tandem with the agent. The x-axis on the plot represents how many times the Canvas Studio agent deployed the underlying plugin code to the test environment for user testing, and the y-axis shows how many times the deploy succeeded (installed successfully and ran without error), meaning the agent cleanly implemented the user's requested changes. In this one plot, we've got doctors, nurses, dietitians, therapists, billers, schedulers, quality directors — taking control of their software to get the exact outcomes they need. It's pretty awesome to see folks push their projects to 40+ deploys of maturity, with the freedom and power to control their software at will instead of it controlling them. If you're not already using Canvas, now is the time to switch so you don't get left behind. Thank you Anthropic for being a great partner in the development of Canvas Studio, and special thanks to the Claude family of models for providing the default power plant that makes Studio run.
-
Beau Gunderson shared thisAdam kindly left out that the LLM spend he's writing about originates primarily from our engineering team, yours truly included. He is right that the DevOps story for internal LLM usage has not been written yet. As engineers we have enormous latitude and broad access to levers that affect our spend tremendously. We can use "/fast" mode in Claude Code to save minutes while diagnosing an outage, for example. We can use Claude code reviews to increase confidence in our work (at a steep price, hence our recent work to build our own review agent at a fraction of the cost). And we can choose between thinking levels and agent workflows to get our work done. Sometimes "/effort max" is warranted. Sometimes multi-agent fan-out for a range of solutions (aka a bake-off) or "/goal"-oriented automated work results in fantastic optimizations. These choices are applied at a personal level... not everyone has my CLAUDE.md or will make the same choices I do. We do surface what works and what doesn't, but the decisions remain with the individual... which means sharing context around good decision-making become critical, for both agents and humans. Tools like skillshare are beginning to provide some of the framework for optimizing usage and ensuring everyone starts from the same page. What's working for your engineering org?Beau Gunderson shared thisLLM spend is the new AWS bill. Not COGS that operates off revenue (this is coming soon), but an operating expense modeled in our unit economics. At Canvas, our LLM spending is now north of 10% of our payroll, and growing. So optimizing it has become mission critical. Claude has become infrastructure that scales with headcount, and we are learning the same lessons we learned about cloud spend a decade ago. Recently Andrew Hines asked us to audit our token spend for low vs. high ROI activity. Three patterns emerged in the data: 💰 The expensive activity isn't humans typing. The top cost driver last week was a background alert dispatch service, not a person. Developer API keys outpaced our heaviest individual users. The cost leverage is in the agents and pipelines, not the seats. 🧠 Reshaping information is cheap, agents are not. The "rewrite this email" workflow people worry about barely moves the needle. The agent that reads a repo, plans changes, and executes against them autonomously is where the dollars go. 💧Spend without an owner drifts. AWS taught us the playbook e.g. reserved capacity, annual commitments, usage dashboards, optimization investments. LLM providers will build the tools for this because customers will demand it, and there’s a win/win in driving stronger customer ROI from each $ spent on tokens. JP Patil already opened that conversation with Anthropic. tl;dr is we are already starting to budget LLM spend against headcount OpEx, give it ownership and attention, build monitoring and alerting tools, think strategically about our contract, and revisit the data on a regular cadence. We now need both FinOps and DevOps for LLMs. The companies that build it early will build compounding equity while the rest pay rent.
-
Beau Gunderson shared thisRecently much of the Canvas team went to San Francisco for our inaugural BUILD event. We invited our customers to come to the Log Cabin at the Presidio to see the bleeding edge from a nearly century-old vantage point. We showed off Studio, a conversational interface for creating Canvas plugins, and Navigator, an agent designed to provide tier 1 support 24/7. Our customers used Studio and the existing Claude Code-based Canvas Plugin Assistant to create plugins and showed off their work to close out the day. We saw nearly 400 successful plugin deploys using Studio and 29 completed plugins. Giving clinicians the tools to build their own workflows leads directly to increased efficiency, better outcomes, and less time spent fighting the EHR. All of our customers want customization. That used to mean a ticket and a wait... now they can ask for and receive it directly from Studio.
-
Beau Gunderson reposted thisBeau Gunderson reposted thisIn a few hours of collaboration, and ~$3K in Claude tokens, here's what our customers created today at BUILD. It's a snapshot of clinical, financial and operational automations across a full spectrum of outpatient care represented in the Canvas customer base. And it's a signal of the power of the Canvas platform and the new Studio interface. You bring your domain expertise, we bring the SDK, and the agent makes it real. Some highlights: 1/ CKD patient ingestion, CHF tier-1 enrollment, and a pre-visit summary card for a value-based kidney care model. 2/ FDA food and medication recall alerts on affected charts, provider state licensure tracking, and an internal knowledge base for nutrition medicine. 3/ A Connective Health registry dashboard for ALS patients alongside a medication reconciliation workflow for dementia care. 4/ Prior-authorization letters from chart context and automated outreach waves to close open care gaps for NP-driven primary care. Before Claude Code (BCC) this is many human-hours of software developer work, months and quarters of roadmap. After Studio (AS) it is a few hours and one day of incredible collaboration with builders and AI-assisted coding. We're living in the future 🚀
-
Beau Gunderson reposted thisBeau Gunderson reposted thisIf you’re a health tech startup trying to get a BAA with Anthropic... 🪦 Here’s what you should know about what’s covered and what isn’t: Things I’ve heard from customers just this week about their experience trying to get a BAA with Anthropic… - “I asked about a BAA, but… they didn’t seem to know anything about it.” - “We had to go to their manager to find a friend of a friend who knew their head of sales to tell them that we’ve been waiting for three months to get a response from their sales team.” To get Claude Code covered under a BAA, you need: Zero Data Retention enabled, an enterprise plan, a qualified account, and the number of a sales rep. That all takes time and $$$, if you can get it at all. Developers building agents on top of Epic and Cerner APIs end up running Claude Code in environments where PHI is present: test fixtures, local database dumps, dev logs. That's all in scope for HIPAA. The terminal can be covered with ZDR configured. But the web interface, desktop app, and review and security tools have no coverage path; they’re incompatible with ZDR regardless of your plan 🥺 We wrote up a guide for digital health teams struggling with this BAA debacle: what's covered with your Claude BAA (if you have one), what requires ZDR and what doesn't, and what you still need to build on your own to remain compliant. Link in comments 👇 Happy to chat in comments or DMs if anyone has more questions that aren’t answered in there 👋
Beau Gunderson replied to a comment
1mo
Matt Wolff amazing! we're in haines today, taking the ferry to skagway tomorrow and will be there until wednesday
-
Beau Gunderson reacted on thisBeau Gunderson reacted on thiswe’re gonna have multi agent swarms communicating via fax to dismantle the healthcare system
-
Beau Gunderson liked thisBeau Gunderson liked thisI want to tell you a story about my brother-in-law. He's a hardworking electrician with almost three decades of experience. Me, I'm a physician and a guy who usually thinks, and often is, smart enough to figure most stuff out. I can spend HOURS on YouTube and with multiple AI tools doing deep research on an electrical problem I have at my home. More often than not, I've spent money on new tools and new equipment to help me troubleshoot the problem. I've torn perfectly good switches right out of the wall and thought it was a good idea (from YouTube) to put some rubber boots on while I tested a rogue breaker. Sometimes, whether I was right or not, the problem ends up fixed and I make sure to congratulate myself. But, most of the time, after hours of digging, I call my brother-in-law. He chuckles, gives me a little smirk and then asks the questions that comes from years of experience and a depth of expertise that quiets the noise and pulls forward the true signal that tells us what is really the problem and how we should approach solving it. I think you get the point of the story. Now, I do want to point out that learning is part of the goal for me. I like to learn. I like to understand. And, although my wife may disagree, those hours spent down rabbit holes investigating electrical mysteries have given me a conversational fluency and level of understanding in electrical work that allow me ultimately to get the most value out of working WITH my expert electrician brother in law. And most crucially, not being fool enough to think I can replace him.
-
Beau Gunderson reacted on thisBeau Gunderson reacted on thisMost people figure when you get the expensive version, you stop being the product. That’s rarely the case. Here’s another one: A restaurant near me went to paid parking a few years back. Next time we were picking a place, my friend said, "Not there, the parking sucks." We never went back. Same food, same everything, and the parking lot was the whole reason. Nobody in that meeting thought they were making a brand decision. They thought they were making a parking decision. Same thing with the screen. Somebody priced ad inventory in the cab, the number looked good, and that revenue lands this quarter. The guy who quietly becomes a Toyota guy instead shows up years later, and by then nobody connects it to the screen. I drive a 2017 Subaru Outback I paid $15k for. My friends call it a soccer mom car. It has never tried to sell me anything. What's the smallest thing that made you quit a brand for good?
-
Beau Gunderson liked thisBeau Gunderson liked thisPhyscopaths billionaires with zero social license to operate. #billionairecancer
-
Beau Gunderson liked thisBeau Gunderson liked thisThis is the AI industry’s equivalent of the Iraq War's “Mission Accomplished” banner. Do you want to know the REAL reason that they're suddenly saying this? It's a coordinated effort to get past the term entirely. They need to diminish the definition of it, and then stake their claim on it. This is also why Altman recently been disparaging of it, calling AGI a "useless marketing term". It's become albatross on their necks (even though it is their reason for existing), so they just declare it has finally happened, and try to never bring it up again. Also good to note that Dario isn't saying anything of the sort; dig into the contention between Dario Amodei and Jensen Huang to see Jensen's OTHER motivation here. As Gary Marcus brilliantly said: if/when AGI arrives, "you won't have to squint to see it".
-
Beau Gunderson reacted on thisBeau Gunderson reacted on thisSomebody cashed out of me last week. No big deal, happens all the time. But this time we tracked it and built a process. Halfway through we hit a flaw that had sat in the mechanism for five years, because I had never audited the full path. Everyone got their money, it was just taking too long! Now we have a semi-automated process...
Experience & Education
Volunteer Experience
-
Software Gardener
The Taskforce
- Present 13 years 4 months
Politics
Projects:
- The Day We Fight Back
- Defund the NSA
Tools:
- call-congress
- social-buttons-server
- thedaywefightback.js
- node-geoip-web -
Software Gardener
Digital Democracy
- Present 11 years 8 months
Environment
I traveled to Peru with other developers to help improve a data collection system for monitoring environmental issues like oil spills and deforestation.
Publications
Languages
-
English
Native or bilingual proficiency
-
Spanish
Elementary proficiency
-
Hatian Kreyol
Elementary proficiency
Recommendations received
12 people have recommended Beau
Join now to viewExplore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content