Re-inventing our craft: how Alan’s Data team is shaping its future with AI
Two Alans and a question
Our company is named after Alan Turing, who laid the mathematical foundations for everything we’re now building on top of, and Alan Watts, who taught that clinging to a fixed identity is the surest way to miss what’s emerging. Both feel relevant right now.
Turing’s question, “can machines think?”, has quietly evolved into something more personal: what should we become when machines can think well enough to help?
We don’t expect AI to give us a shorter workweek. What we’re after is something different: the chance to shape what our role becomes, to own more of our impact end-to-end, and to turn a craft we already love into one we could love even more.
We don’t know exactly what we’re becoming. But Watts would remind us that the answer isn’t to grip tighter to what we were, but to stay curious about what’s emerging. So that’s what we’re doing: exploring actively, for a chance to shape the way forward.
The scale is different this time
Tools have always been evolving in data. New databases, new orchestrators, new BI platforms, each one an incremental improvement to an existing process. LLMs and the agentic frameworks built on top of them are just another tool in that lineage, yet the scale is different: where most tools improved one part of the workflow, agents potentially allow us to work differently on all parts of the stack, simultaneously.
And crucially, these changes compound: when coding is cheaper, and data navigation is trivial, and visualization is near-instant, the cost of building an MVP collapses. Your framing becomes your first prototype. You can implement an end-to-end pipeline in a fraction of the time, try it for real, break it, and rebuild with stronger convictions. That feedback loop (from idea to working system to learning) used to take weeks. Now it can take hours.
Some of what’s already changed for us:
- Prototyping replaces planning. What used to require a team and a sprint can now be tested in a day. We explore by building, not by debating.
- Execution shifts to agents; judgment stays with us. Agents write the first draft, build the dashboard, run the query. We focus on intent, correctness, and trade-offs that require human discernment.
- Self-serve analytics expand. The bottleneck shifts from “who can write the query” to “who has the right question and can find the right data.”
Expanding the craft
This is the shift that matters most to us.
One of our leadership principles is “Mission is the boss”: everyone acts as an owner of Alan. In our data team we hire full-stack profiles to solve ambiguous problems, not implement pre-specified solutions.
Agents make these principles more tangible than ever. They lower the barrier on skills that used to require deep specialization, so one person can own more of the full picture: identifying the problem, writing the software, running the analysis, automating the workflow, synthesizing the knowledge that accelerates a decision. As a concrete example, many in our data team now contribute directly to our product’s backend code-base.
At the same time, data is democratizing across Alan. Product managers, insurance specialists, and ops are editing our data assets and building new ones. A senior PM recently ran a full product engagement analysis in two hours with our AI-powered data tools. Out of ~60 users of our data IDE, ~25 are outside the data team, using it almost daily.
People are autonomous on their project where before they would need a Data specialist. The talent stack has collapsed: one person with the right agents can now cover ground that used to require a team of specialists. This shifts our role from bottleneck to enabler.
So what’s left of the data craft?
Framing the system, making the right trade-offs, and shaping the infrastructure that both agents and non-specialists can exploit: well-documented datasets, semantic layers, privacy controls, evaluation frameworks. We don’t know exactly what our craft will look like next year, or five years from now, but we can actively shape it rather than wait and see.
What we’ve tried and learned
We’ve been experimenting broadly, with Nao and Claude Code for data development, Dust and Nao Agents for analytics agents, and early explorations of semantic layers. Everyone in our team has a mandate to explore and try.
Here’s what we’ve found so far:
- Navigating your data graph is trivial for agents. Tell it “I don’t understand why this table has this issue” and it will trace it back to the upstream dependency and the specific record causing you trouble. Hours of debugging, collapsed into a prompt.
- Building visualizations is a delight. The first draft is often near-perfect, and a quick prompt with the few tweaks you need is enough to get to a final version. No more clicking everywhere within the limits of your BI tool, or deploying a custom JS-based stack.
- Contextualization is key. Formal semantic layers, thorough documentation, validated queries or any other artifact that adds context to your data beyond the code that produced it can make a big difference in how well agents can leverage your data.
- Security and privacy are the hardest trade-off. The more data agents can access, the higher the quality of their answers. That means we need strong controls over how agents interact with your data. Our members trust us with some of the most sensitive information there is, and our innovation must not create new risk to that trust.
- Agents need guardrails beyond security. Even without sensitive data at stake, agents with write access will make a mess: aggressive git pushes, unexpected tables in your database, every conceivable unit test including ones that check 1 + 1 = 2. Clear guidelines and scoped permissions are essential.
- Trust is the bottleneck. Agents can produce answers fast, but if every answer requires a manual check before you can act on it, the productivity loop breaks. We need infrastructure that makes agent output trustworthy by default: data contracts, consistency checks, evaluation datasets…
The uncomfortable questions
This is not all unadulterated excitement. It comes with tough questions, and being honest about them is part of shaping this responsibly:
- Privacy and security don’t stop at our own controls. Every new capability that touches data is also a new surface for risk. The more your agents can do for you, the more harm a malicious actor can do with the right credentials and a single prompt.
- Beyond security, there are ethical questions about the impact of AI on our environment and society.
- There is fatigue from the increased volume of code to review, generated documents to read, and the overall acceleration in the pace of content production.
- There are legitimate questions about whether we’re using a bulldozer where a gentle shove would do.
- There are questions about the value and authenticity of human interactions when our emails, Slack messages, and other written contributions are AI-generated, even with human input.
We don’t have clear answers to most of these questions. But we’d rather sit with them while actively shaping the outcome than discover the answers passively.
Why we’re doing this
Our culture helps make this possible. Alan is built around an ownership mindset, experimentation, and learning from failure. Our leaders lead by example, starting with our co-founders. And we have champions: people who naturally carve the time, share their explorations, and make the new way of working feel possible for everyone else.
Does it create tension with business priorities? Yes. It slows you down temporarily. But the investment compounds: you either extract value directly, or you learn to learn the next one faster.
Turing gave us the foundations. Watts would tell us to stay curious about what’s emerging. We’re trying to do both: build on solid ground, and stay open to the craft becoming something new.
We’ve decided to seize the tools and invent the role for ourselves: to grow the range of things we can do, to refine how we do them, and to increase our impact on Alan’s mission of transforming our healthcare system(s). Not because we’re certain where it leads, but because the opportunity is both too great and too perilous to just bear witness.
If any of this resonates and you want to join our exploration while impacting the healthcare journey of our members, we’re hiring!
