I wrote a new blog post with @pydantic.dev! With the Pydantic AI + DBOS integration, agent runs are automatically checkpointed to Postgres. Executions are durable and observable with DBOS + Pydantic Logfire. Check it out: pydantic.dev/articles/pyd...

A presentation so nice, I'm doing it twice. I'll be delivering my Building Reliable Workflows in Java talk again on Feb 24th, this time in Seattle. If you're a Java (or Kotlin!) developer, come hear all about what you can do with DBOS. www.meetup.com/seajug/event...

OpenAI Agents SDK + @dbos.dev integration is now live! You can run OpenAI agents with durable execution backed directly by your database. DBOS embeds as a library, so it works anywhere, even inside a local coding agent. Check it out, and any feedback is welcome: openai.github.io/openai-agent...

DBOS is coming to skills.sh You can now teach your coding agent (Claude Code, Codex, Antigravity, anything) how to write durable workflows by just running: `npx skills add dbos-inc/agent-skills` Available in Python and TypeScript.

Check out this new release of Pydantic AI: parallel tool execution in DBOS agents with deterministic replay. The newly added execution mode runs tool calls in parallel, then emits events in a stable order. This makes agent execution reproducible and crash-recoverable without sacrificing latency.

Love this template. It shows how you can build reliable apps in a serverless environment with DBOS.

Cool new collaboration: a @vercel.com + @supabase.com + DBOS template! It's a simple pattern for running reliable background jobs: - Write tasks as DBOS workflows (regular TS/JS) - Enqueue them from your app - Vercel functions pick them up and run them automatically vercel.com/templates/te...

Harry Pierson speaking at the Portland Java User Group #dbos @dbos.dev @devhawk.net

Fivetran - DBOS webinar tomorrow, Jan 28 at 9AM PT/12PM ET Learn how to build a control plane that safely manages Fivetran Connector SDK state changes using DBOS durable workflows. It enables idempotency, automatic recovery, and enterprise-grade reliability‚ without complex custom orchestration.

A nice feature in the latest DBOS Python release: you can now call async steps concurrently with native syntax from a workflow. To make that possible, we added code to set a unique ID for each step before it begins, so even though the step executions overlap they’re checkpointed deterministically.

The January DBOS User Group kicks off 2026 tomorrow: luma.com/v94dyuk5 Agenda: ​- How to self-host and scale DBOS Conductor. ​- New DBOS Conductor pricing. ​- CommonSense AI explains their DBOS use case and journey, transforming local government operations with AI. - 2026 roadmap See you soon!

This is a must-read annual review on database systems. 👀

Our first meetup of the year! 🍻

The keyword of 2025 for me is: community. This includes @dbos.dev, @southbaysystems.xyz, and several open-source projects. 🧵 Thanks to everyone who was part of my 2025. I'm proud of what we built, clearer about what matters, and excited about the year ahead.

2025 has been a wild year for DBOS. Huge thanks to you all for the support, feedback, and trust along the way. Looking forward to making reliable software even easier to build in 2026 🚀

Simplicity wins. I really appreciated the shoutout: "Integrating durable workflows through DBOS solved our distributed system headaches without adding ops overhead." www.decodingai.com/p/building-v...

People used to complain that the DBOS UI wasn't available locally. Good news: it is now, and it's free for open-source local dev. Run a local DBOS UI right on your laptop and actually see your workflows while you build. Debugging hits different when it's visual. docs.dbos.dev/production/s...

Cool new feature: explicit queue listening! You can now configure exactly which queues each of your workers listens to. This is especially helpful if you have heterogeneous workers (such as some workers for GPU tasks and others for CPU-intensive tasks) as you can easily divide work between them.

Want to hear what two database professors Mike Stonebraker and @andypavlo.bsky.social really think about AI and data systems, plus a recap of 2025 and predictions for 2026? In case you missed the webinar, watch the full recording here: www.youtube.com/watch?v=RnQK...

What happens when applications are written almost entirely by LLMs? @andypavlo.bsky.social explains what’s already happening, and why future enterprise apps may start with vibe coding.

Ever wondered how AI teams avoid burning cash every time they run an image model?

When you combine patching + versioning, you get: - Multiple workflow versions for large refactors (blue/green deployment). - Inline hotfixes for long-running workflows that need to be corrected. Would love feedback or war stories from anyone who's wrestled with workflow versioning in production.

I was once skeptical about workflow patching, mostly because of all the horror stories I heard from other developers. But this implementation looks nice: a patch point is just another workflow step. That shift makes the patching interface more intuitive than I expected. docs.dbos.dev/python/tutor...

Goblins moved from external workflow services to DBOS for faster execution, a unified stack, and full control over debugging and uptime. Alp said "when your database is essentially one millisecond away from your servers, every workflow is just blazing fast.”

Looking forward to this! Fun fact: Andy generated those Christmas-themed headshots himself (pretty sure via Nano Banana).

Good memories from #QConSF Conversations about system design tradeoffs, and someone told me they learned about SKIP LOCKED from my talk. That's why I love QCon: the technical exchange, not product pitches. I now have my second Postgres elephant plushie thanks to @yugabytedb.bsky.social

For Ontologize, reliably spinning up enterprise software environments at scale required durability and simplicity. DBOS let them scale their deployment pipeline with Python and zero rearchitecting.

In the upcoming webinar, we’re hosting a conversation with two leaders in the database world, Mike Stonebraker and Andy Pavlo. They will discuss the trends most likely to influence the software you build and the architectures you choose. 🗓️ Weds, December 10, 2025 | 1 PM EST

Happy Thanksgiving! Our dev is already shipping code before dinner.

As teams move toward agentic AI, systems need to handle long running, independent tasks with very little supervision. The challenge is keeping these workflows stable, consistent, and easy to recover when something goes wrong.

Big hugs to friends and family working overtime this time of the year to keep systems alive during the surge. I'm reminded how many people quietly keep the world running. Thank you for everything you do!

Thanks so much for hosting the podcast! Really enjoyed our conversation.

Yesterday was our final @southbaysystems.xyz talk event of 2025 (what a year!). Huge thanks to @julianhyde.bsky.social and Yuanyuan for their awesome talks! A big shout-out to our amazing host, @startreedata.bsky.social, for the venue and food!

Youtube: youtu.be/h4FrkIf2Poo Spotify: open.spotify.com/episode/1SxH... Apple podcasts: podcasts.apple.com/us/podcast/e...

The latest episode of the #BugBash podcast has @qianli.dev from DBOS talking about the hard truths of ergonomics. As always, if it ain't easy, it probably won't happen. Full episode links below.

Gunnar is such an incredible writer! He explains durable execution better than I do. I love seeing more lightweight, database-backed approaches getting attention, and really appreciate the thoughtful shout-out for @dbos.dev. Always grateful for folks who push the space forward.

Durable execution is underrated. In real systems, any step can fail at any time. Peter breaks down why reliability is such a challenge.

Build durable, exactly-once workflows in your favorite language. DBOS Go ✅ DBOS Java ✅ DBOS Python ✅ DBOS TypeScript ✅ docs.dbos.dev

Here is a nice article summarizing the talk (thanks stiller.bsky.social) 🔗 www.infoq.com/news/2025/11...

Thanks QCon for having @qianli.dev and @jedberg.net. Their talk covered how durable computing and DBOS Transact boost reliability, remove heavyweight workflow coordinators, and reduce both complexity and cost.

Building reliable AI systems is hard. In this video, @petereliaskraft.net walks through a deep research agent that many of us want to build and the kind that fails in all the surprising ways. Failures are inevitable and a durable workflow is what will make your agent production ready.

I will be at #QConSF next week and co-presenting with @jedberg.net! We're speaking on Monday at 11:45 am on "Compiling Workflows into Databases: The Architecture That Shouldn't Work (But Does)". DM me if you want to catch up or grab coffee. 🔗 qconsf.com/presentation...

80% of backend code is retry and error-handling logic. With durable execution, that disappears. Every input, output, and error is saved, giving you full observability and debugging out of the box.

We're back on HN! This time with our Postgres-backed durable execution library for Java. It plugs into existing Spring apps with minimal changes and gives you durable workflows, queues, retries, and scheduling without extra infrastructure. Feedback very welcome! github.com/dbos-inc/dbo...

Build durable, exactly-once workflows in your favorite language. DBOS Go ✅ DBOS Java ✅ DBOS Python ✅ DBOS TypeScript ✅ docs.dbos.dev

When Yutori looked into Temporal, the cost was clear. It would mean rearchitecting their workflows. With DBOS, they could orchestrate durable workflows using just a few Python decorators and a Postgres database.

Moreover, their execution is fully observable in the DBOS console (workflow graphs + management, shown in the screenshot) and instrumented with Logfire to trace token usage and cost per step. Link to the original PR: github.com/pydantic/pyd...

I added two new demos showing durable agent patterns (parallel vs sequential) with @pydantic.dev AI + @dbos.dev: - Deep Research: large fan-out parallel workflows - Twenty Questions: long sequential subagent chaining Both are long-running, so durable execution is necessary to survive crashes.

Great sample app from the dbos team of an inbox for durable agentic workflows- group executions by pending/approved/denied, with state all stored in a PostgreSQL database. https://github.com/dbos-inc/dbos-demo-apps/tree/main/python/agent-inbox