Armon Dadgar

Really interesting interview with @martin_casado and @vishalmisra (CS prof) on the Bayesian nature of LLMs and what that implies. Great if you want to build an intuitive understanding!

It's fun to be back in SF and see the energy and spark in the city again. Very noticeable the change from a year or two ago. City feels more vibrant, safer, and cleaner.

This video is a masterpiece

@vivekramaswami Thanks for sharing! Agreed both can be wrong/overcorrections. Long term, question is if SaaS can maintain 80%+ margins when cost of building software drops by 100x.

@armon Agreed! x.com/vivekramaswami…

It’s interesting to see the extreme disconnect between late stage private valuations and public software companies. Privates still doing 10-20x revenue while public companies dropping to 2-4x.

The Gell-Mann Amnesia of using AI is real. On topics you know better, you can call BS and mostly the response agrees the initial answer was non-sensical. But then we blindly trust the output on domains where we lack subject matter expertise...

@erratlinezero @DirtFishRally You raise an excellent question...

@armon @DirtFishRally Why on earth was hashiconf never colocated with rally??

First time Rally driving at @DirtFishRally, had an awesome time. Very different than being on a track, but can't complain when you get be sideways through every turn!

I'm an openly gay founder of a very successful company, who lived in the Bay Area for 9 years. This article is completely insane, and at best represents a small microcosm of the Valley.

Agent skills help agents use your products, build in your codebase and enforce your policies. They’re not just words - they are what the unit of software for agentic devs, and need powerful dev tools to match. That is what @tessl_io offers. Tessl is the package manager and development platform for skills. It offers a full dev lifecycle, helping you generate, evaluate, distribute and observe skills & context, developing them to the professional grade they warrant. Today, I’m excited to announce the general availability of our task evals, which help you understand how good your skills are. Such insight is critical to making your skills great, avoiding regression, and applying learnings from their real world usage. For example: @Cisco's software-security skill shows a 1.8X improvement in securing coding in its benchmark, and @ElevenLabs's agents skill boosts success by almost 3X! However, not to name names, we often see skills that provide minimal uplift while consuming context window space, or even degrade functionality. As Spencer Kimball, CEO of Cockroach Labs, put it when we shared early versions of this: evaluation is what makes agentic coding outcomes converge instead of drifting. Task evals are joining a long list of powerful context development tools, such as: * Review skills against quality best practices * Generate and maintain skills and docs for using your libraries & platform * Distribute versioned skills to your dev team and ecosystem * Consume skills easily and safely, and keep them up-to-date Skills are a central part of software development. If you’re serious about making agentic dev successful in your org, or helping your customers’s agents use your products, you need to invest in them. We hope Tessl can help. Check out links in the thread to get started!

I am making Claude Code read research papers related to problems we are solving before writing plans for @tensorlake's compute engine (think about it as Hashicorp's Nomad/Apache Mesos but for distributed sandboxed agents). Ex - Claude wrote an implementation of container lifecycle management in the dataplane(a component like the Kubelet of K8s, Nomad Agent, Mesos Executor, etc) without an explicit state machine - the result was a mess with a lot of race conditions in creating, terminating and reporting state of containers. Made it read a few papers I read myself back in the day- Omega, Borg, Apache Sparrow and boom it understood how state machines and log-based events help with building robust systems that manages lifecycle of a resource. It then re-wrote the code and it was - 1. Less lines of code 2. No race conditions 3. Improved observability of capturing state transitions. 3. Correct state transitions - such as killing a container which is still starting up but the server wants it gone. Make your agents learn about computer science the same way you would to build invincible software!

AI eliminated the natural barrier to entry that let OSS projects trust by default. People told me to do something rather than just complain. So I did. Introducing Vouch: explicit trust management for open source. Trusted people vouch for others. github.com/mitchellh/vouch The idea is simple: Unvouched users can't contribute to your projects. Very bad users can be explicitly "denounced", effectively blocked. Users are vouched or denounced by contributors via GitHub issue or discussion comments or via the CLI. Integration into GitHub is as simple as adopting the published GitHub actions. Done. Additionally, the system itself is generic to forges and not tied to GitHub in any way. Who and how someone is vouched or denounced is up to the project. I'm not the value police for the world. Decide for yourself what works for your project and your community. All of the data is stored in a single flat text file in your own repository that can be easily parsed by standard POSIX tools or mainstream languages with zero dependencies. My hope is that eventually projects can form a web of trust so that projects with shared values can share their vouch lists with each other (automatically) so vouching or denouncing a person in one project has ripple effects through to other projects. The idea is based on the already successful system used by @badlogicgames in Pi. Thank you Mario. Ghostty will be integrating this imminently.

Recently switched to @obsdmd for note taking. Having notes saved as Markdown lets me use Gemini or Claude CLI to interact with the notes. Seamless cross-device sync via iCloud too. Has been a huge improvement over using Apple Notes.

Had a great time speaking with the latest batch of @DubHacks on launching and growing a startup. @UW campus really shined for the occasion!

Huge congrats to the @getmetronome team! They've built an awesome product for consumption billing, and @Stripe is a fantastic home.

Interesting read from @EpsilonTheory on the massive spending in AI, $4-5T over the next four years (equivalent to WWII) and the crowding out of investments across the economy. epsilontheory.com/world-war-ai/

#HashiConf last week was the 10th anniversary. Lots of great announcements, including TF Stacks GA, TF Actions and Search Beta, Vault SPIFFE enhancements, Boundary RDP support, and Project Infragraph our real-time graph database of infrastructure assets: hashicorp.com/en/blog/10-yea…

Really annoying that @Apple lets apps send you a geo-fenced notification when location is restricted to only while in use. Lyft, Uber, etc. all abuse this which feels like a privacy violation. developer.apple.com/documentation/…

Surprise! I made a surprise return for the 10th anniversary HashiConf. Even most of the employees didn’t know. It was really fun to see everyone and congrats on all the announcements.

Attending @HashiCorp’s HashiConf today as it celebrates its 10th year. I helped organize the first one when I was 29. It’s really special to see a lot of the elements and programs I helped create still live on today. 🥹 @armon 🥰