Smaller Teams, More Agents: Orchestrating AI Across a Legacy Enterprise

A four-person team just rebuilt something that used to take eight people two quarters — in one. Now multiply that across an enterprise, add agents to every team, and the question stops being “can we go faster?” It becomes “can we orchestrate all of this without losing the plot?”

That’s the operating reality Sharon Hunt navigates at JLL — a 200-year-old commercial real estate enterprise running on a sprawling, largely acquired technology stack, now rolling out AI across the entire organization. Her world is property management: the Building Engines platform and its flagship app, Prism, plus the broader suite JLL’s own property teams depend on. It’s enterprise scale meeting one of the least tech-forward industries there is — and AI is forcing every layer of it to move at once.

In this episode of Accelerate Podcast, host Becky Flint, Founder and CEO of Dragonboat, sits down with Sharon for an honest look at what AI velocity actually demands at enterprise scale: how teams reshape when agents join them, where speed creates new bottlenecks, and why none of it works without a shared source of truth underneath.

• Speed turned out to be the easy part. A small team rebuilt JLL’s entire tenant portal front end in a quarter on an LLM-optimized design system — work that used to take six to eight people far longer. Pair that with agents filling seats (a five-person squad becomes three-plus-agents, and 30 teams can become 60 or 70) and the hard problem stops being output. It’s coordination: keeping dozens of teams, humans, and agents pointed at the same outcomes.
• Agents and skills are becoming the product, and build-vs-buy is being redrawn live. Some workflows turn into production agents, like matching incoming payments to invoices across $1.2B in North American transactions, with a human kept in the loop for compliance. Others become reusable skills — CAM reconciliation, lease-renewal analysis — that get property managers 90% of the way and adapt property by property. The new question: can an AI plus a skill close the gap before you procure another enterprise tool?
• The data underneath is the real blocker. Sharon’s sharpest point doubles as her most unsolved: the app layer is almost secondary. The actual work is pulling fragmented data like leases, contracts, tenant sentiment stuck in SharePoint, inboxes, and local drives into something agents can reason over. Commercial real estate has no shared definition of even “a building,” so without unified identifiers and clean relationships, agents infer and trip. She’s building a semantic layer she calls the Property Management Intelligence Hub, and is the first to add: just getting started.
• Legacy code slows the agents down, too. A codebase that’s hard for a human to follow — twenty hand-rolled components for one form field is hard for AI to move through fast. Her working theory: engineers become mini-dev managers whose real job is grooming the codebase so agents can run cleanly. She’s careful to call it a theory, not a fix.
• Outcomes, not output, are what keep the speed honest. JLL is moving from feature factory to an empowered model with three KPIs — save users time, create trackable value (NOI), and increase deployment speed. If a piece of work doesn’t move one of them, it shouldn’t get built. AI accelerates the building; clear outcomes are what keep all that velocity aimed at value instead of activity.
• Role lines are blurring, and the people winning aren’t precious about it. Engineers show up with ideas, designers ship code, PMs ship code. Sharon sees a new “builder” role emerging — though product taste, knowing what’s actually worth building, still decides who’s effective. The ones holding tight to the old org-chart boundaries are the ones feeling the squeeze.
• Ship fast enough and the jam moves downstream. Once an enterprise can produce change at speed, the constraint relocates: can support, docs, sales collateral, and demos keep pace, and do customers even want that much change? In a change-resistant industry, Sharon isn’t convinced enablement can keep up, and worries clients get worn out by the churn.

Agentic enterprises run on live product portfolio context. As Sharon’s story shows, AI speed only compounds into value when humans and agents share one trusted, up-to-date source of truth. That’s where Dragonboat connects the dots: giving teams and their agents the shared context and memory of their product operating reality, so everyone can decide, orchestrate, and act toward outcomes — not just ship faster.

Reference

• Sharon Hunt on LinkedIn: https://www.linkedin.com/in/sharonehunt/
• Becky Flint on LinkedIn: https://www.linkedin.com/in/beckyflint/