The two AI workflow tools your team depends on just merged

THE AI BRIEF

Today's signal: Prefect acquired Dagster. The two biggest open-source orchestration tools just merged, and the stated reason is that neither could handle agentic workflows alone.

In today’s issue:

• Main story: The two biggest AI workflow tools just merged into one company

• Also worth knowing: DoorDash published agent architecture with 24% higher conversions, GPT-5.6 ships Programmatic Tool Calling, 16 Nobel laureates signed a joint AI labor warning, and more

THE READ

Prefect acquired Dagster. If your team runs data pipelines or is building AI agents on top of them, this changes the vendor landscape.

Prefect announced Monday that it is acquiring Dagster, its closest competitor in open-source data orchestration. Both are tools companies use to schedule, monitor, and retry automated workflows: the kind of infrastructure that sits underneath AI agents and makes sure things run reliably when a model calls an external tool, hits an error, or needs to pick up where it left off.

Jeff Lawson, Prefect's CEO, framed this as the foundation for what he calls "agentic orchestration," the problem of automating software that is itself autonomous. Dagster and its commercial tier, Dagster+, will continue as distinct products and brands. The people who built Dagster are staying. Both open-source communities remain active.

The reason this matters is not the consolidation itself. Two mid-sized orchestration vendors combining into one is a routine M&A story. What matters is why it is happening now. Prefect's announcement explicitly names agentic systems as the reason for the deal: traditional orchestration tools were built to run known sequences of steps, but agents don't follow known sequences. They branch, retry, call tools dynamically, and take paths no one anticipated when the workflow was written. The existing market leader in orchestration, Airflow, which powers most enterprise data pipelines, was not built for this. Neither Prefect nor Dagster individually had the scale to credibly claim they could.

The combined company is profitable, according to the announcement, which gives them runway to make the agentic infrastructure bet without needing to raise capital in a market where infrastructure valuations are under pressure.

A 15-minute assessment that turns your answers into a real decision artifact: your readiness level, your six-axis shape, where you're strongest, where you're constrained, and what not to build yet.

My read: if your team is building anything agent-based, whether that's internal tools calling APIs, AI workflows touching customer data, or anything running unsupervised, you are going to need orchestration tooling sooner than you think. What I keep hearing in conversations with teams that are actually shipping agents is that orchestration is the part no one planned for. They get the model working, they get the tool working, and then the question becomes: what governs this thing when it runs at 3 am and something unexpected happens? The fact that two of the three serious open-source options just became one company doesn't close the market. It signals the market is clarifying. Airflow's incumbency is not threatened this week, but the competitive pressure on its roadmap just increased.

SPONSORED BY CLUTCH

Hire secure AI teammates that work 24/7.

Hire pre-built AI teammates. Give your engineers and operators a platform to ship their own AI apps. Stop losing sleep about what is running where.

Clutch is the platform behind both: pre-built agents for the workflows your ops team should automate first, plus the integration plane your team's vibe-coded apps and Claude Code projects plug into. One platform. Real production. Visible and safe by default.

Built for ops, engineering, and security teams that are tired of the shadow-AI surface area inside their own company.

ALSO WORTH KNOWING

DoorDash reports 24% higher conversions from its AI shopping assistant. DoorDash published its architecture for Ask DoorDash, which blends language models with specialized sub-agents, MCP tooling, and a memory layer. Early results include up to 24% higher checkout conversions and 17% larger baskets. This is one of the first major consumer companies to publish both the architecture and the business metrics for a deployed vertical agent: worth reading if you're building a case for agents internally.

OpenAI's GPT-5.6 launches Programmatic Tool Calling for agent control. GPT-5.6 ships with Programmatic Tool Calling in the Responses API, which makes agent tool use a programmable API-level primitive rather than something you configure through prompts alone. GPT-5.6 Sol also sets a new benchmark on Agents' Last Exam with a score of 53.6, beating Claude Fable 5 by 13.1 points. For teams evaluating which frontier model to build agents on, the efficiency gap between GPT-5.6 Terra and Fable 5 at roughly one-sixteenth the cost is the number worth stress-testing.

Sixteen Nobel laureates join economists calling for AI economic preparation. A coalition including Daron Acemoglu, Joseph Stiglitz, Paul Krugman, and Simon Johnson published a joint statement at wemustactnow.ai calling on governments and institutions to prepare for AI's labor and economic disruption. The signatories also include figures from Anthropic, OpenAI, and Google. It is an advocacy statement, not a policy proposal, and carries no binding force: but when 16 Nobel laureates sign something together, board conversations about workforce planning tend to follow.

Imbue open-sources mngr for parallel multi-agent runs. Imbue released mngr, a free open-source tool for running multiple AI agents in parallel across different models simultaneously. Practical for teams that want to compare model outputs on the same task without writing custom orchestration code. Another sign that multi-model workflows are becoming the default architecture assumption, not the advanced use case.

P.S. If your team is building on top of data pipelines and you want to think through where orchestration fits in an agent architecture, hit reply. I'm running a few of these conversations right now, and the patterns are worth sharing.

Back tomorrow,
Haroon