After the Data Intelligence Engineer Comes the Outcome Engineer
MotherDuck argues that the new data job is the data intelligence engineer: a person who automates ingestion and modeling by running AI agents. I think they have the direction right and stop one step short. As agents take over the engineering, including the running of other agents, the job that remains is the outcome engineer: one person who owns a business number, like a small P&L, and is judged on moving it, not on which tools they know.
Here is their argument as fairly as I can put it, where I agree, and where I think it goes next.
What MotherDuck argues
In The dawn of the Data Intelligence Engineer, MotherDuck CEO Jordan Tigani names a role he first proposed at the company's Data Outpost conference on November 4. His starting point is that the work a data person does with AI no longer fits the existing titles.
In his framing, the data intelligence engineer takes over the data engineer's job (ingestion) and the analytics engineer's job (transformation and modeling) "by creating and conducting software agents". They write what they know about the data into shared context that other agents use. The goal is self-service: anyone who asks a question should get an answer as good as the one the company's best analyst would give.
His example: One of MotherDuck's data engineers moved the team to a new Airflow version over a weekend, with minimal supervision. That used to be a week of work.
Where MotherDuck is right
The old titles are breaking. Data engineer and analytics engineer split the work by layer of the stack. When one agent writes the ingestion code and the transformation model in the same session, that split stops meaning anything.
Context is the work. We see the same in our own deployments: agents rarely fail on SQL. They fail on meaning nobody wrote down, and accuracy decays when that meaning goes stale (we wrote about that in Your AI Analyst Will Be Worse Next Month). Writing down what the data means is the highest-leverage thing a data person can do this year.
Self-service is finally real. For twenty years, self-service BI meant a drag-and-drop tool most people never opened. An agent that answers in plain words, with its work shown, changes who can ask.
If you run a data team, the data intelligence engineer is a better job description than the one you posted last year.
One step further: who conducts the conductor?
The weekend Airflow upgrade is the most important detail in the post, and I read it the other way around. A week of work became a weekend with little supervision. The work shrank, and so did the supervision. Nothing says the second trend stops.
Conducting agents is engineering. It means writing context files, choosing tools, writing evaluations and fixing a run that went wrong. Those are the tasks agents are improving at fastest, because they are text, code and tight feedback loops. Agents already draft context from documentation and past questions, and propose a fix when someone flags a wrong answer. Today a person reviews every change. Next year the review gets lighter, and the year after an agent does most of it.
So a role defined by the technology it runs is a stage, not a destination. Data engineers became conductors of ingestion agents. Conductors become reviewers of agents that conduct agents. Each step moves the human one level up, and each level has less engineering in it.
What does not move is the top of the stack: deciding what to optimize, defining what done looks like, choosing between trade-offs, and answering for the result. Agents will propose goals and model trade-offs, perhaps better than we do. They cannot be accountable for them, because accountability is a promise a person makes to other people.
Self-service is also a means, not an end. Nobody's quarterly goal is that people can ask questions. The goal is lower churn or higher margin. Once every question gets a great answer, the bottleneck moves from answering questions to choosing them and acting on what comes back.
What is an outcome engineer?
An outcome engineer is responsible for one business outcome, with a target and a date, and is judged on whether the number moved. Think of a small P&L: net revenue retention for mid-market accounts, gross margin on one product line, or time to first value for new customers. The tools are their choice. The result is not.
"Engineer" stays in the name on purpose. The work is still systematic: they build the loop of data, agents, experiments and follow-ups that moves the number, and they can read the SQL when a figure looks wrong. What disappears is the need to build every part of that loop by hand.
It is not a manager title either. They do the work, with agents doing most of the execution.
What an outcome engineer does all week
- Writes the definition of done. Does retention mean revenue or logos, which cohorts, measured on which date? That paragraph is the spec every agent works from.
- Asks instead of building. Instead of a ticket for a pipeline and a dashboard, they ask which accounts will likely churn next quarter, and why. The answer comes back in minutes with the query behind it, and they check the one number that looks too good.
- Makes the trade-off. An agent can model a 10% discount for at-risk accounts against the margin it costs. Choosing is the job, and so is explaining the choice.
- Runs the experiment, and kills it. They launch an intervention, have an agent report on it every Monday, and stop it when the number does not move. Stopping your own project is harder than starting one.
- Audits the evidence. When an agent names a cause, they ask what else could explain it. They approve changes to shared definitions, because a quiet change to "active customer" moves everyone's numbers.
- Answers for the number. On Friday they explain in plain words why it moved, to the person who holds them to it.
None of this requires knowing Airflow. All of it requires knowing the business.
What this means for data teams and hiring
Organize by outcome, not by layer. Instead of an ingestion team, a modeling team and a BI team, you get one owner per number, working on a shared platform. The platform team gets small. For the next few years, that platform is where the data intelligence engineer does their best work.
Hire for judgment and domain. Swap "how would you build this pipeline?" for "tell me about a number you moved, and how you knew it was you and not the season." Stack knowledge has a short half-life now. Knowing why customers leave does not.
If you are a data person, attach yourself to a number. Pick the metric your work already feeds, learn the business behind it, and ask to own it. We gave the longer answer to "will AI replace data analysts?" in an earlier post.
This is also the person we build Dot for. Dot is an AI data analyst: an outcome owner asks it a question in Slack, Teams or the browser and gets the answer with the SQL behind it, instead of waiting for a pipeline. The data team's job becomes keeping the definitions it answers from true.
The job named after the result
Job titles named after a technology last as long as the technology does. Data intelligence engineer is a good name for 2026, and MotherDuck deserves credit for saying out loud that the old titles are finished. But the agents that build pipelines this year will be conducting each other next year. The job that lasts is named after the result: pick the number, define done, make the call, and answer for it.
Rick Radewagen
Rick is a co-founder of Dot, on a mission to make data accessible to everyone. When he's not building AI-powered analytics, you'll find him obsessing over well-arranged pixels and surprising himself by learning new languages.