In our Creator Highlight sequence, TDS editors chat with group members about profession paths, writing, and sources of inspiration in information science and AI. Right now we’re happy to share our dialog with Mike Hulse.
Mike is a know-how chief working on the intersection of information engineering, AI, and structure, serving to organizations rework complicated information environments into dependable, easy-to-use programs. He has a powerful full-stack background and designs end-to-end options that stability technical depth and enterprise worth. Alongside our consumer work, we construct and share sensible instruments and insights round information platforms, AI programs, and scalable architectures.
Do you take into account your self a full-stack developer? How does your expertise throughout the stack (from front-end to database) change the best way you view the information scientist position?
I do this too, but it surely doesn’t suggest I construct each layer personally. To me, full stack means understanding how architectural choices at one layer form system habits, danger, and value over time. That perspective is important when designing programs that want to face up to change.
This angle additionally influences how I view the position of a knowledge scientist. Fashions created in notebooks are only the start. The actual worth comes when these fashions are constructed into operational programs with the appropriate information pipelines, APIs, governance, and user-facing interfaces. Knowledge science has a better influence when handled as a core half of a bigger system, fairly than as an remoted exercise.
You cowl a variety of subjects. How do you determine what to concentrate on subsequent? And the way have you learnt when there is a new matter value contemplating?
I are inclined to observe repeated friction. After I see a number of groups fighting the identical downside, whether or not technical or organizational, I take it as a sign that the issue is structural fairly than remoted and value addressing at an architectural or course of degree.
We additionally deliberately experiment with new applied sciences, not for novelty, however to grasp their trade-offs. A subject is value writing about if it solves an actual downside that we presently face or reveals a danger that isn’t but broadly understood. Lastly, I’ll write about subjects that I personally discover attention-grabbing and value exploring. It’s because for those who keep , you possibly can dig deeper.
I wrote about LangGraph, MCP, and self-hosted brokers. What do you assume is the most important false impression individuals have about AI brokers in the present day?
Brokers are really highly effective and open up new prospects. The misperception is that they’re easy. Right now, it is easy to assemble a cloud infrastructure, join an agent framework, and create one thing that appears prefer it’s working. This accessibility is efficacious, but it surely hides a whole lot of complexity.
When brokers transfer past the demo, the actual challenges floor. State administration, permissions, value administration, observability, and failure dealing with are sometimes underestimated. With out clear boundaries and possession, company operations change into unpredictable, costly, and dangerous. These are extra than simply prompts with instruments. These are long-lived software program programs and should be designed and operated accordingly.
in your article Layered structureSo, you talked about that including performance typically seems like “open-heart surgical procedure.” For inexperienced persons or small information groups who need to keep away from this, what’s your high recommendation on establishing the structure?
“The one fixed is change” is a cliché, and for good cause: Optimize for change, not preliminary supply velocity. Even minimal layered considering helps to separate area logic, software stream, and infrastructure considerations.
The objective is to not have an ideal structure or good classification from day one. It is about creating clear boundaries for the system to evolve with out fixed rewriting. Just a little self-discipline up entrance goes a great distance as your system grows.
We performed a benchmark PostgreSQL insert technique And he stated, “Quicker is just not all the time higher.” What situations would you deliberately select a slower, safer insertion methodology in a manufacturing ML pipeline?
When accuracy, traceability, and recoverability are extra essential than uncooked throughput. For a lot of pipelines, saving just a few seconds of execution time is of little profit in comparison with the dangers posed by weak ensures.
For instance, pipelines that feed regulatory reporting, monetary decision-making, or long-term coaching datasets profit from transaction safety and express validation. Silent information corruption is rather more pricey than accepting modest efficiency tradeoffs, particularly when the information turns into a long-term asset that others construct upon.
inside you private agent assistant On this article, we constructed a 100% non-public, self-hosted platform. Why was avoiding “token prices” and “privateness leaks” extra essential to you than utilizing a extra highly effective cloud-based LLM?
In my every day work, I’ve skilled that trusting the system is the idea of system implementation. Token prices, opaque information flows, and exterior dependencies subtly have an effect on how the system is used and perceived.
We additionally made a acutely aware selection to not route private or delicate information by means of exterior cloud suppliers, as there are restricted ensures about how the information can be handled over time. By protecting the system self-hosted, we had been in a position to design an assistant that’s predictable, auditable, and according to European privateness expectations. Customers have full management over what the Assistant can entry, reducing the barrier to make use of.
Lastly, not all use circumstances require the biggest or most costly mannequin. By separating the system from a single supplier, customers can select the mannequin that most accurately fits their necessities, balancing options, value, and danger.
How do you assume the every day work of information professionals will change in 2026?
Regardless of frequent stereotypes, information and software program engineering is a extremely social career. I am a powerful believer that a very powerful a part of the work is finished earlier than you write the code. This implies coordinating with stakeholders, understanding the issue area, and designing options that match inside present programs and groups.
This upfront work turns into much more essential as agent-assisted growth accelerates implementation. With out clear objectives, context, and constraints, brokers improve confusion fairly than productiveness.
In 2026, information professionals will spend extra time shaping programs, defining boundaries, validating assumptions, and making certain accountable habits in manufacturing.
Waiting for the remainder of 2026, what do you assume are the massive subjects that may outline 2026 for information professionals? Why?
Generative AI and agent-based programs will proceed to develop, however the greater change can be maturation into first-class manufacturing programs fairly than experiments.
That transition depends upon trusted, high-quality, accessible information and sturdy engineering practices. In consequence, full-stack considering and system-level design will change into more and more essential for organizations that need to apply AI responsibly and at scale.
To be taught extra about Mike’s work and preserve updated together with his newest articles, observe him on TDS or linkedin.
