Every investment management firm I speak with right now is either actively experimenting with AI or feeling the pressure to do so. While the AI productivity case is undeniable, a critical gap is coming up more and more  in these conversations – one that matters enormously in a regulated, high-stakes environment like investment management.

How to operationalize AI: The gap between a good idea and a production-grade workflow.

The pattern I see playing out repeatedly is where a quant or portfolio manager has a genuinely powerful idea, prototyped using Python, a spreadsheet, or built with an AI assistant. It works brilliantly in a sandbox, the output is compelling and leadership gets excited. But then it stalls.

Because moving from “this works on my laptop” to “this runs reliably across our entire platform, for multiple users, with auditable inputs and consistent outputs” is not a small step. It’s an entirely different problem.

In most industries, that gap is an inconvenience but in investment management, it’s a risk. You’re dealing with fiduciary obligations, regulatory scrutiny, client money, and processes where a poorly governed model can cause real harm – to portfolios, to compliance standing, and to trust.

AI makes the prototype phase faster than it has ever been, that’s the good news. The bad news is that it makes the governance gap more visible, not less. When you can generate ten new ideas in the time it used to take to develop one, the bottleneck shifts entirely to: how do we get any of these safely into production?

This is why I believe that the firms that will win the AI race in investment management are not necessarily the ones generating the most ideas. They’re the ones that have built a harness – an infrastructure layer that lets them move from idea to prototype to production systematically, with the guardrails that the industry demands baked in from the start.

That means having answers to the questions a CTO or Chief Compliance Officer will inevitably ask: Who authorized this workflow? What data did it touch? What did it cost to run? Can I audit what it did and why?

These aren’t nice-to-haves. In a regulated environment, they’re the price of entry for any AI capability that graduates beyond a proof of concept.

Long before “AI” became the dominant conversation in investment management, Jacobi was focused on a single, unglamorous but critical problem: how do you take the proprietary logic that lives inside an investment firm (whether it is in a portfolio manager’s head, a spreadsheet, or a bespoke model) and move it into infrastructure that is reliable, scalable, and accessible across the entire organization and even by clients?

From day one, that meant private deployment infrastructure, APIs, and a software development kit designed specifically for investment managers. And it meant embedding our own investment engineers directly inside client teams. Engineers who understood both the software and the investment logic well enough to convert one into the other. More broadly speaking, other industries are catching up to this model: what Palantir popularized as “Forward Deployed Engineers” is something Jacobi has quietly been doing in investment management for over a decade.

The AI moment hasn’t changed that thesis. It has validated it.

The launch of Jacobi’s new AI Workbench and AI Production Layer is the natural evolution of that same infrastructure — designed to give investment teams the environment where they can develop proprietary IP quickly, with the governance and guardrails that make it production-ready rather than permanently trapped in a prototype. The value isn’t in any AI model itself. The value is in the harness: the structured, auditable, compliant infrastructure that makes AI reasoning reliable inside real investment workflows.

Model-agnostic by design, we’re not competing with OpenAI or Anthropic – we’re making them useful inside your investment processes. Jacobi provides the deterministic, auditable substrate that grounds that capability in your firm’s actual data, entity model, and workflow logic.

The firms that are moving fastest right now aren’t the ones who’ve found the best AI model. They’re the ones who’ve figured out how to industrialize what their people are already building – compressing the journey from idea to prototype to production from quarters into weeks.

The AI race in investment management won’t be won by firms with the best models. It will be won by firms with the best workflows.

The parallel I keep coming back to is what Palantir built for defence and enterprise operations: not a model, not a point tool, but a platform that made AI actionable inside the workflows where decisions actually get made. Investment management needs the same thing and the firms that recognize this early will have a durable structural advantage over those that are still assembling point tools and hoping they add up to something. The intellectual rigor is there, the ideas are there, the AI capability is there. What’s often missing is the infrastructure layer that converts all of that into production workflows the whole organization can trust and build on.

If you’re an investment management firm navigating this moment — whether you’re deep into AI adoption or just beginning to ask serious questions – I’d love to compare notes. The firms getting this right are solving a very specific operational problem, and it’s one worth talking about openly.

Tanya Bartolini is Chief Revenue Officer at Jacobi Strategies, the investment workflow platform purpose-built for OCIO’s, asset and wealth managers.