Steve Novak, Definian: "AI Is the Worst It Will Ever Be Right Now — But That Is Exactly What Should Keep You Sharp"

The third episode of The Process People features Steve Novak, head of AI and automation practice at Definian, a firm that helps organisations get more out of their data through data governance, data engineering, migration, analytics, and agentic automation. Steve's specific focus is on the hardest part of the AI journey for most organisations: not building a pilot, but making it to production. Getting agents deployed, maintained, and trusted at scale is a different problem from demonstrating that AI can do something impressive in a controlled setting.

The conversation with Alain Cohen of Famla AI covers the framework Steve uses to approach any client problem before reaching for a solution, what genuinely keeps him up at night about how people are using AI tools, where he thinks AI is still misunderstood on both sides of the hype curve, and why he considers critical thinking the single most important capability a practitioner can develop right now.

Who Is Steve Novak and What Does Definian Do?

Steve Novak runs the AI and automation practice at Definian, where his work focuses on helping organisations move beyond experimentation with AI and into production-grade agentic systems. That transition is harder than it looks: it requires getting the data right, designing the semantic models accurately, engineering the agents correctly, and managing the expectations of stakeholders on both ends of the hype spectrum — those who believe AI will automatically solve complex problems and those who remain sceptical that it can do what practitioners claim at all.

Before leading this practice, Steve built his perspective across consulting engagements in data and analytics. His current focus on agentic AI puts him at the precise intersection of where the technology is moving fastest and where organisations are most likely to get it wrong.

The Framework Steve Novak Uses Before Reaching for Any Solution

Asked what he would add to the standard playbook for his discipline, Steve Novak's answer had nothing to do with a specific AI tool or methodology. It was a question framework — and a deliberately simple one.

Before any project begins, before any solution is scoped or any technology is selected, he insists on getting clear on three things: what the client or organisation is trying to solve, why they are trying to solve it, and who is affected by or accountable for the outcome. Only once those three are understood clearly does the conversation turn to how. And his position on the how is direct: it is the least important piece of any project or objective.

This is a stance that will resonate with anyone who has watched a well-resourced technology implementation fail because nobody agreed on what problem it was actually solving. The how — the tool, the platform, the methodology — is substitutable. The what, why, and who are not. Getting them wrong at the start means everything that follows is optimising for the wrong outcome.

How Steve Novak Structures His Week: Deliver, Sell, Build

Asked what a typical week actually looks like in practice, Steve Novak described a structure that he considers fundamental to how a consultant should operate regardless of the specific discipline. Three things, every week, without exception.

The first is to deliver great work — serving current clients, making sure commitments are met and that the quality of what is delivered matches or exceeds what was promised. The second is to sell great work — actively looking for new opportunities, talking to new people, understanding what problems others are facing and where Definian can help. The third is to build great teams — meeting with individual team members, understanding their challenges and aspirations, removing blockers, and ensuring that the people around him have what they need to succeed in their own careers.

He is explicit about where the leverage lies in this model: team building is the foundation. A strong team is what makes the other two pillars — delivery and business development — sustainable at scale. Building people who can deliver and grow independently is not a secondary priority that gets attention when client work is slow. It is the mechanism through which everything else compounds.

What Keeps Steve Novak Awake: The Intellectual Atrophy Risk of AI Adoption

Asked what is keeping him professionally awake at night, Steve Novak did not point to a capability gap or a technical limitation. He pointed to a behavioural risk that sits inside the organisations successfully adopting AI tools.

As AI handles more of the analytical, generative, and decision-support work that practitioners previously did themselves, there is a real risk that people stop exercising the cognitive muscles that made them effective in the first place. Coding instincts blunt. Business judgment becomes shallower. Content produced by AI is accepted without the scrutiny that would have been applied to human-produced content. He calls this intellectual atrophy — and it is the concern of someone who has seen how quickly capability degrades when it is not exercised.

The productivity gain is real and near-term. The cost is long-term and harder to see until it has already accumulated. A team that has handed over its thinking to a machine for two years is not the same team it was before, and reversing that is not as simple as switching the tool off.

Steve Novak on AI's Current Limitations: Usually the Prompt, Not the Model

Alain Cohen asked whether there are areas where AI has consistently disappointed or failed to deliver what Steve expected. His answer was striking for someone who works in AI professionally every day: not really.

What he observes is that when AI produces bad output, the root cause is almost always upstream of the model itself. The context provided was insufficient. The semantic model underlying the data was misaligned. The prompt lacked the guardrails needed to constrain the output usefully. In his experience, the failure modes are fixable through one of these layers — they are rarely evidence that the model cannot do what was asked of it.

He made a point that frames the entire conversation around AI capability: AI is the worst it will ever be right now. Every day it becomes marginally more capable. That trajectory creates its own challenge — if today's limitation is not a structural one but an input quality one, then the practitioner's job is to get better at providing the right inputs, not to wait for the model to improve. The model will improve regardless. The practitioner's skill at working with it is what determines how much of that improvement they can actually capture.

Steve Novak's Advice for the Next Generation: Be Curious, Show Up, Help Others, and Think Critically

Asked what advice he would give to someone who wants to build a career in AI, automation, and transformation consulting, Steve Novak returned to the same principles he opened the conversation with — and added one more that he considers increasingly essential.

Be curious: about what others are working on, about what will make the people around you successful, about areas that sit outside your current expertise. Show up: be present, be available, be willing to engage with things that are outside your comfort zone. Help others be successful: the returns from investing in other people's success compound back in ways that are hard to trace but consistently real.

And the fourth, which he added specifically because of where AI is taking the profession: think critically. Build and maintain the capacity to evaluate information, challenge assumptions, and arrive at independent judgments. As machines take over more of the execution layer — the code generation, the content production, the data synthesis — the practitioner who can still think independently about what matters, what is true, and what to do will be disproportionately valuable. The task doer is increasingly automatable. The problem solver is not.