About
Founder and CEO of Petabridge, a company dedicated to helping .NET developers build…
Activity
3K followers
-
Aaron Stannard shared thisI've been wanting to get this video out the door like at least 6 weeks: "Introducing Netclaw: Always-on, Autonomous Agent Powered by Akka .NET" Why have I spent hundreds of hours working on it, how does it help me, how does it work, and how can Netclaw help you? Netclaw is in the same class of AI assistant as something like OpenClaw or Hermes Agent, with some key differences: * Simple, secure, and reliable - we embrace the principle of "fewest moving parts" and have an approvals system designed to prevent agents from making irreversible or destructive changes without your express approval. This system applies both to shell execution and MCP connections. * Scaled to local inference - we've done all of our evals on Qwen3.5/3.6:35b-a3b primarily, but I've also run Netclaw no problem with models as small as Qwen3.5:9b. * Chat-first - designed to be run entirely through Slack / Discord / etc, apps you already use every day. We include rich UI elements that are rendered in those applications for things like approvals, markdown tables, and more. * Reminders and webhooks - used to drive LLM automations either on a schedule or in an event-driven fashion. Still subject to your shell approval system and we include prompt injection scanning for skills, webhooks, and so on. * Sub-agents - you can define custom sub-agents or use our stock ones to help manage context windows, just like you would with Claude Code or OpenCOde. * Skills and the Netclaw SkillServer - we support skills, invocable skills, and ship a separate stand-alone skill-server application you can use to serve skills that comply with the AgentSkills.io standard and our feed is exposed using the CloudFlare Agent Skills RFC. This makes it feasible to sync all of your organization's skill definitions + resources from a single location controlled by you. * Web search, web fetch, and browser integration - we support multiple search providers And it's all open source, licensed under Apache 2.0, written in C#. Check out the video below to learn more. https://lnkd.in/gc4Pbi34Introducing Netclaw: Always-on, Autonomous Agent Powered by Akka.NETIntroducing Netclaw: Always-on, Autonomous Agent Powered by Akka.NET
-
Aaron Stannard posted thisWe did this earlier this year, but I was overdue for a blog post / video on it. We removed FluentAssertions from all of Akka .NET's public TestKit packages earlier this year in response to their license change. I don't begrudge OSS maintainers for trying to get paid for their work, but my specific vision for Akka.NET is that it acts as a stable foundation for building the world's most important software. I believe that in order to do that Akka.NET must be permissionless to use. I explain in the video / blog post below how, even with FOSS exceptions to commercial licenses, these types of dependencies conflict with our mission. https://lnkd.in/gMJEuSFh
-
Aaron Stannard shared thisTonight at NHDNUG: "Should You Build an MCP Server?" by Christopher Bremer, Ph.D Doors open at 5:30pm at Beach Walk Coworking. Come join us for free pizza and some in-person learning. Petabridge is sponsoring! https://lnkd.in/gEEC-YPD
-
Aaron Stannard shared this🔭 End-to-End Distributed Tracing for Akka.Streams Join us tomorrow (Wed May 13) for the May Akka.NET Community Standup, where we're going to walk through one of the biggest observability features we've shipped in a while. Akka.NET 1.5.66 introduced a framework-owned `"Akka.Streams"` ActivitySource that propagates Activity context end-to-end through stream graphs across async stage boundaries, fan-in merges, and fan-out broadcasts. The result: an unbroken distributed trace from `akka.msg.recv` → `akka.stream.ingress` → `akka.stream.stage` → your user `SqlClient` / `HttpClient` spans. We'll do a short live demo showing fan-in ActivityLinks when concurrent orders merge into a single outbound batch. Also on the agenda: 🛡️ Akka.NET 1.5.65 cluster sharding lease coordination fix (the ~6-minute shard unavailability bug on Kubernetes) 🔒 New `allow-unregistered-types` serialization hardening setting ⚠️ Important upgrade notice — anyone on 1.5.66 with a third-party persistence plugin needs to move to 1.5.67 #dotnet #AkkaNET #OpenTelemetry #DistributedTracing #Observability 📺 https://lnkd.in/g9amegwv 🕛 12pm CDT (1pm ET / 10am PT / 17:00 UTC)End-to-End Distributed Tracing for Akka.Streams - Akka.NET Community Standup 5-13-2026End-to-End Distributed Tracing for Akka.Streams - Akka.NET Community Standup 5-13-2026
-
Aaron Stannard shared thisNew Blog Post / Video: "Managing Long-Running Tasks Inside Akka.NET Actors" I've answered this question in various forms over the years for customers that were building transcoding systems, ETL pipeline, data-intensive applications, etc where long-running background jobs (minutes / hours, not seconds) and it was time to finally make a video. We've had a number of users sign up for TextForge since I announced it and some of these folks, myself included, have ~15+ year old gmail accounts with 100k+ emails in them - and some of our users choose to sync ALL of that. That's a lot of metadata that has to be synced with the Google APIs, thousands of API calls that all have to be retried potentially. In this video + blog post I explain how we do that reliably with a time-tested and relatively simple Akka.NET pattern: https://lnkd.in/gkJrKkJS
-
Aaron Stannard shared thisA side project I've been working on for ~5-6 months that I'm happy to share: TextForge. I built TextForge originally to solve my own problems as a small business owner: I have a ton of things to do and unfortunately dealing with email junk consumes at least 2-3 hours per day. Most of this is urgent but not strategically important: sales follow-ups, passing redlines back and forth, filling out vendor onboarding paperwork, and dealing with other random requests ("can you resend us that receipt?") "I use Claude Code and OpenCode extensively for programming and we rely on approval workflows to prevent LLMs from turbo-nuking our code and production infrastructure. Why can't I do the same for email?" Enter TextForge - it's an MCP server or HTTP API (does both) that allows you to choose what you want an LLM to read, allows an LLM to compose rich emails with signatures + attachments, and _won't send anything unless you approve it_. If Claude muffs an email and drafts something that is either factually wrong or makes you sound like kind of an idiot, you can edit it yourself or ask Claude to do it for you. Nothing gets sent until you click "approve" I've been using this for months originally as a self-hosted application. Decided to SaaSify it after watching OpenClaw users get.... surprised... by what agents decided to send on their own volition. It's helped me: 1. Meet my sales quota 2 months into the quarter 2. Fill out numerous pieces of tedious paperwork 3. Keep Pipedrive fully synced 4. Successfully negotiate our vendors down thousands of dollars Right now we only support Gmail, but I've planned IMAP and Outlook / Office365 access. Let me know what you think in the comments. https://textforge.net/
-
Aaron Stannard shared thisDon't miss Lara Rivera and Dave Rivera's presentation on building, Grimmloch - an MMO powered by Akka.NET on Wednesday. You can watch it live here or on YouTube: https://lnkd.in/gFm4qrGQAaron Stannard shared thisThis month's Akka.NET Community Standup features a Show & Tell from David and Lara Rivera of Opulent Code, presenting Grimmloch - a persistent text-based MMORPG for roleplayers with the entire backend built on Akka.NET. Topics covered: - Akka.NET v1.5.61 and v1.5.62 releases (critical cluster sharding fixes, VectorClock corrections, logging stability) - Phobos v2.11.0 with first-class log-trace correlation - Akka.Management v1.5.61 Azure coordination fix - Grimmloch Show & Tell: Akka.Cluster.Sharding, Akka.Persistence, and DistributedPubSub working together in a real production game server - Akka.NET v1.6 and v2.0 roadmap updates Agenda: https://lnkd.in/gqpwJ6dF Grimmloch on Steam: https://lnkd.in/gNJFDSvc Discussion Thread: https://lnkd.in/gu7f4utG Akka.NET Home: https://getakka.net/ Learn Akka.NET: https://learnakka.net/ Akka.NET Discord: https://lnkd.in/gepMcDeV Akka.NET Support: https://lnkd.in/gNagih2r
-
Aaron Stannard shared thisI'm giving a talk tonight at the AWS Houston Meetup: "Escaping the Cloud Complexity Tax with Software Ideas from the 1970s" Key theme is that statelessness leads to a ton of unnecessary coordination + concurrency complexity in many modern web / cloud applications because no one really owns anything in those transient transaction-based designs - and there's a better way to do it and we've known it since the 70s.Escaping the Cloud Complexity Tax with Software Ideas from the 1970s, Thu, Feb 26, 2026, 5:30 PM | MeetupEscaping the Cloud Complexity Tax with Software Ideas from the 1970s, Thu, Feb 26, 2026, 5:30 PM | Meetup
-
Aaron Stannard shared thisModeling LLM conversations as durable events in Akka.NET's persistent actors - that's how I'd do this in my own applications. Another great post from Lara on her MMOs built on top of Akka.NETAaron Stannard shared thisThree LLM calls, $0.06 spent, 8 seconds elapsed. Call #4 fails. Do you restart from the beginning or resume where you left off? Traditional approaches restart from step 1. Event sourcing resumes at step 4. The Problem AI-powered game systems run multi-step workflows that span multiple LLM API calls. An NPC backstory might require: Call #1: Generate character concept ($0.02, 2 seconds) Call #2: Develop personality traits ($0.02, 3 seconds) Call #3: Create relationship history ($0.02, 3 seconds) Call #4: Weave narrative coherence... fails LLM APIs fail. Networks drop. Nodes crash. Rate limits hit. These aren't exceptional disasters—they're operational reality. Without durable state, one failure at step 4 means restarting from step 1—repeating three successful API calls because one failed. For a single NPC, that's annoying. For hundreds of NPCs running autonomous workflows across a distributed cluster, it's unsustainable. Event Sourcing Changes The Equation Akka.NET Persistence doesn't store workflow state as a mutable object. It stores the history of what happened as immutable events: WorkflowStarted NodeExecuted (call #1 complete) LLMResponseReceived (response cached) NodeExecuted (call #2 complete) LLMResponseReceived (response cached) NodeExecuted (call #3 complete) Call #4 fails When the actor restarts, it replays those events. The workflow sees: "I completed steps 1-3. Resume at step 4." No duplicate API calls. No wasted money. No repeated latency. The same pattern applies to GOAP planning (expensive A* pathfinding through action-state space) and RAG retrievals (database queries with vector similarity search). If the computation was expensive, event sourcing ensures you never repeat it unnecessarily. Beyond Cost Savings Event sourcing also provides a complete audit trail. When debugging why an NPC made a specific decision two hours ago, we can replay the exact sequence: which LLM prompts were sent, what responses came back, how long each step took, what tokens were consumed. That's impossible with mutable state snapshots. Once the state updates, the history is gone. The Architecture Our LLMAgentActor persists events to PostgreSQL via Akka.Persistence.Sql. Each workflow execution becomes a durable event stream that survives failures, cluster rebalancing, and actor migration between nodes. Cluster Sharding ensures each workflow runs on exactly one node at a time. If that node fails, the cluster supervisor restarts the actor on a surviving node. Event replay reconstructs exact state, and execution continues. The Bottom Line When AI operations are expensive, you can't afford to repeat them. Event sourcing turns failures from restarts into resumptions—with zero duplicate costs and complete execution history. #AkkaNet #EventSourcing #DistributedSystems #AI #GameDev #LLM
-
Aaron Stannard liked thisAaron Stannard liked thisA voice agent, orchestrated in F#. You talk to it, interrupt it mid-sentence, and ask it to edit the page. It does it by voice, through tool calls into live browser state. In the screenshot, it just wrote my shopping list into the notes box. One F# codebase targets the browser via Fable and coordinates the backend pieces, with a shared wire protocol so the frontend and backend stay in sync. Same source, CPU or A100-backed demo. My takeaway: F# is seriously underrated for AI orchestration. When things get async, stateful, and tool-heavy, the type system earns its keep. Closed source for now. Rough, but satisfying. I’ll keep the demo online for a few days. Feel free to try it, interrupt it, or have a slightly weird conversation with it if you’re bored. Demo link is in the comments. Please be gentle, it is a temporary shared demo with short sessions. Video to follow. #fsharp #dotnet #AI #LLM
-
Aaron Stannard liked thisAaron Stannard liked thisThis is the entire write-side of an event-sourced document system. Two screens. Registration is the architecture. The domain is one pattern match — create vs. edit, wrong-id routing, saga verdicts, all of it exhaustive and checked by the compiler. FCQRS: functional CQRS + event sourcing on Akka.NET, in F#. The example is Focument. I'm teaching this hands-on at DDD Europe — 11 June, Antwerp. Empty composition root to a working system with a saga and a resumable projection, live. If you've wondered what event sourcing looks like with values and functions instead of objects and mutation, come build it with me. #fsharp #dddeurope #eventsourcing #dotnet
-
Aaron Stannard liked thisAaron Stannard liked thisWhat an incredible few days at NDC Copenhagen 2026 last week 🙌✨. I had the pleasure of representing Twilio at the historic Øksnehallen venue in the heart of Copenhagen’s Meatpacking District. This event was absolutely fantastic, bringing together around 600 software engineers, architects, and tech leaders. Some of my favourite highlights and key takeaways from the week: 🤖 Navigating the AI Hype Cycle: The overall tone of the conference was heavily anchored in where AI is actually going. Sessions from tech legends like Richard Campbell and Emmz Rendle tackled everything from data center energy limits to the realities of "vibe coding." The biggest consensus? Software engineering isn't going anywhere, we just need to build responsibly and use the right tools. 🛠️ Elise Kristiansen did a fantastic talk on why accessibility matters and how you can easily make your websites, talks, and presentations far more accessible. And Aaron Stannard talked about all the lessons learned when it comes to making architecture decisions. 🎨 The AI Anime Hook: I brought back our interactive AI anime-artwork creator app, and it was a total showstopper. 40 unique attendees, creating over 275 artworks. I used it during my talk and it was the perfect bridge to showcase our live console tools, Ngrok tunneling, and the Twilio Sandbox. 💡 Community Spark: One of my favorite moments was hearing from an attendee, who was inspired by my talk, and now wants to build a custom Raspberry Pi-based photo frame at home for his kids' parties, returning themed transformations. This is exactly why I love doing what I do! 🥹 🧦 The Swag Tax: It's official, our Twilio stickers remain absolute gold, and my personal stash of scrunchies was completely raided. So glad you loved all the things. The NDC Conferences crew also did a fantastic job: giving us custom printed hoodies and totes (I spent a bit of time prepping my hoodie, so thanks to Alan for making it happen), making lots of cool stickers, and creating a conference vibe that held to lots of great interactions. I'm only sad I missed the padel-ball tournament! 🎾 I would have had a great time running around the court; definitely signing up for that next time. A massive thank you 🙏 Jakob Bradford, Henriette Motzfeldt, Heather Fabela, Alan G. Frausto, Steinar Braaten, Kjersti Sandberg and Emily Svea Holweck for all your hard work and great creativity. To the AV crew and the ever talented photography and videography team, you're all amazing. Until next time, Copenhagen ❤️. #NDCCopenhagen #Twilio #DeveloperRelations #TechCommunity #AI #SoftwareEngineering
-
Aaron Stannard liked thisAaron Stannard liked thisStir Trek 2026 was great this year and the most recent conference I attended since... Stir Trek 2025! For me, this year felt so different because of how much my life and career has changed just over one year. I'm working with a new, wonderful, and brilliant team talking new and complex challenges together, which has required me to learn and adapt much more than in recent years. I found myself reflecting on this a lot during and after the conference. Some talks that I greatly enjoyed included. - DevOops: a satirical guide on doing DevOps Wrong by Victor Frye. The humor ran deep since I experienced a lot of those anti patterns first hand! - The Rewrite from He**: a candid walkthrough of the pain of a project gone wrong and what it actually took to turn it around from Aaron Stannard. - Managing the human Side of Technology: I love coding and development, but truthfully I'll always find deep joy in helping people learn and be their best, and this one really connected with me. Thank you for the stories and examples Jeff McWherter! Stir Trek 2026 also brought to my attention I have not spent much time networking or learning outside of work and expanding my knowledge in the great tech space. It felt like one of those moments you realized you're hyper fixated on the trees and forgetting about the forest. Overall though it was a great way to spend a Friday! Of all the movies at the end to choose from, I went with Galaxy Quest at my teams recommendation! #Stir Trek Conference, Inc.
Experience
-
-
-
-
-
-
-
-
-
-
-
Education
-
Vanderbilt University
-
-
Activities and Societies: Sigma Nu Fraternity, IEEE Student Chapter, ACM Student Chapter, Versus Magazine, VandyStudents.com
Minors: Mathematics, Managerial Studies: Information Systems
Projects
-
NBench
- Present
See projectCross-platform performance benchmarking and testing framework for .NET applications.
NBench is designed for .NET developers who need to care about performance and want the ability to "unit test" their application's performance just like XUnit or NUnit tests their application code. -
Akka.NET
Akka.NET is a C# port of the popular Akka distributed actor system (written in Scala.) I worked with Roger Alsing to port the bulk of it to C#. I'm primarily responsible for Akka.NET's remoting and clustering capabilities.
Other creators -
-
CaptainObvious
-
See projectA simple portal for getting feedback on ideas -mostly an excuse for me to teach myself MongoDB
Honors & Awards
-
Microsoft MVP for Visual Studio and Development Technologies
Microsoft
Microsoft Most Valuable Professionals, or MVPs, are technology experts who passionately share their knowledge with the community. They are always on the "bleeding edge" and have an unstoppable urge to get their hands on new, exciting technologies. They have very deep knowledge of Microsoft products and services, while also being able to bring together diverse platforms, products and solutions, to solve real world problems. MVPs make up a global community of over 4,000 technical experts and…
Microsoft Most Valuable Professionals, or MVPs, are technology experts who passionately share their knowledge with the community. They are always on the "bleeding edge" and have an unstoppable urge to get their hands on new, exciting technologies. They have very deep knowledge of Microsoft products and services, while also being able to bring together diverse platforms, products and solutions, to solve real world problems. MVPs make up a global community of over 4,000 technical experts and community leaders across 90 countries and are driven by their passion, community spirit, and quest for knowledge. Above all and in addition to their amazing technical abilities, MVPs are always willing to help others - that's what sets them apart.
-
Microsoft MVP for Visual Studio and Development Technologies
Microsoft
Microsoft Most Valuable Professionals, or MVPs, are community leaders who’ve demonstrated an exemplary commitment to helping others get the most out of their experience with Microsoft technologies. They share their exceptional passion, real-world knowledge, and technical expertise with the community and with Microsoft.
https://mvp.microsoft.com -
DataStax MVP for Apache Cassandra
DataStax
The DataStax Cassandra MVP Program recognizes key individuals who have gone above and beyond to share their knowledge and expertise to help the Cassandra community learn about the massively scalable NoSQL database.
http://www.planetcassandra.org/mvps/
Recommendations received
2 people have recommended Aaron
Join now to viewOther similar profiles
Explore more posts
-
Zi Dong
Everything is a File: Why Claude Cowork’s Abstraction Matters Anthropic recently released Claude Cowork, and its underlying architecture is worth noting: coding agents are treated as general-purpose agents, and the workspace is abstracted as a filesystem. This approach is particularly interesting to me. During my time at Databricks building a distributed filesystem, we found that files remain one of the most durable abstractions for developer productivity. They provide a standardized interface for: 1. Version Control: Leveraging existing tools like `diff` and `git` for auditing. 2. Scalability: Files allow for straightforward programmatic manipulation in parallel. Agents know how to use battle-tested command line tools. 3. Persistence: They are cost-effective to store and easy to reproduce across environments. We’ve seen this pattern drive significant innovations in storage. Big data frameworks like Delta Lake, Apache Iceberg, and even serverless PostgreSQL variants have moved toward treating object storage (files) as the primary storage layer, separating it from compute to achieve better elasticity and reliability. The Implications for Agentic Workflows: By adopting a filesystem abstraction, Claude Cowork generalizes filesystem manipulation by treating it as the primary interface for AI action. It brings the rigor of developer workflows—like branching and state persistence—to a broader audience, allowing anyone to manage digital projects with the same predictability and auditability we’ve long enjoyed in engineering. Full details here: https://lnkd.in/guTeZ2Gc
2 Comments -
Paul Chiusano
I thought this was a good and very reasonable talk by Eleanor Millman on how to prioritize projects for a platform engineering team: https://lnkd.in/gaYtUHRH A few insights I took from it: - Unlike product features which can be more directly tied to revenue, platform engineering work is a couple steps removed. But that doesn't mean it's unimportant! Far from it, platform engineering projects can make everyone at the company more efficient, able to produce higher quality software, etc. - While "urgency" is always going to be a factor, you don't want to just be putting out fires, you want be able to prioritize work which is "high impact" ... even if that impact isn't felt immediately. - Good rule of thumb: prioritize "highest impact for lowest effort". - The talk has some ideas on what factors to choose for impact, and how to blend them. For instance "speed of development" is one factor, "cloud cost optimization" might be another. The weighting of different impact factors can change over time, depending on the needs of the business. I would say that a lot of companies don't have much methodology here but I can really see the value in codifying it. You can always change the methodology or the weighting if it's spitting out results that don't pass the smell test or you really feel it is leading the company astray. Clarity can give the org more freedom to put resources behind projects that would otherwise never be taken on. When things are unclear, prioritization still happens implicitly, but the decisions tend to be a lot more random and fear-based and the org is worse off as a result.
3 Comments -
Ali Raza
A Unity WebGL build I received was 1.2 GB. The project itself was not complex. It was an interactive survey experience for breast cancer awareness. The problem was how the content was delivered. The entire application contained dozens of decision-based videos stored directly inside the Unity project. Every user path triggered different video files, which caused the WebGL build to explode in size. This created three problems: 1. Very slow loading times 2. High hosting costs 3. Poor user experience for survey participants Instead of trying to compress everything inside Unity, I redesigned the delivery architecture. What I changed a) Removed all video assets from the Unity build b) Moved video delivery to Cloudflare R2 CDN c) Replaced local video references with dynamic URLs d) Added fallback logic for delayed or failed video loading e) Built a lightweight Google Apps Script + Google Sheets logging system to track participant decisions Result Build size reduced from 1.2 GB → 68 MB The experience loaded dramatically faster, hosting costs dropped, and the client was able to deploy the project to their research website successfully. Sometimes performance problems are not about optimization inside the engine. They are about architecture decisions around the engine. Full breakdown: https://lnkd.in/df79C9uj
4 Comments -
Max Millien
Multi-AI workflows in action 🤖🤖 I just published “Two AIs Walk Into a Repo: My Hybrid Workflow with Claude and Jules” — a hands-on look at how I use both Claude and Jules(and also Kilo Code) in my development pipeline. The hybrid setup helps me automate mundane tasks while keeping human oversight for the tricky parts. Explore how I: - Assign roles - Implement guardrails so neither agent goes off track - Integrate their output into code reviews and merge workflows 👉 Read it here: https://lnkd.in/eJN-Ut4q #AI #AgentWorkflows #DevOps #SoftwareEngineering #MLOps #AIinDev #Automation #Claude #Jules #KiloCode
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content