Thought Leadership

What real AI adoption looks like from the inside

The session deliberately avoided the AI conferences’ most tired genre: the slide about what AI will do in five years. Four members of the SPNX team ran live demonstrations of AI use cases already doing billable work.

The entry point was Microsoft Copilot running inside Excel against a client trial balance. In the time it usually takes to brief a junior, the tool produced variance flags, recalculated key ratios, and drafted a usable management-letter structure — all of it inside the same Microsoft 365 tenancy the firm had been paying for since 2023.

One level up, the team demonstrated a Copilot Studio build: a policy and regulatory chatbot trained on firm-specific handbooks and jurisdiction-specific compliance content, serving a Manchester tax team querying UK rules and a Dubai team querying DMCC free-zone provisions from the same underlying architecture, with every interaction contained inside the firm's Microsoft environment.

The third demo crossed the line from tools into agents — AI that takes action across systems, executing sequences of operations against a stated goal. IT in a Box, a Copilot Agent module inside SPNX's Praxis 2.0 platform, provisioned a new joiner's full working environment (hardware, software stack, access controls, security policies) end-to-end, collapsing onboarding from days to hours with a human approving at the control gates.

The fourth demo was the furthest along the curve: two enterprise applications built on Anthropic's Claude and running in production for named clients — an enterprise risk management platform inside India's largest logistics operator, and a vendor and dealer risk system across largest OEM. Both have been live long enough to close books against them.

The purpose of running all four in sequence was to demonstrate a single architectural fact: a firm can begin at Level 1 today and progress to Level 2 by the end of the quarter without rebuilding anything. The continuity argument matters more than any single capability in the demonstration, and it is the argument most firms miss when they evaluate AI tools in isolation.

The foundation? Microsoft Copilot

The question every decision maker gets stuck at is the same. Where does the firm actually start? In 2026, the answer is narrower than the market's noise suggests - Microsoft Copilot. The structural case for starting with Copilot is difficult to beat honestly.

Your firm already runs on Microsoft 365. Copilot lives inside that tenancy. Audit working papers, tax files, and client correspondence never cross the perimeter your IT and risk committees have already approved, which closes the single conversation that kills most AI pilots before it begins — where does the data go, who can see it, what is the retention posture. The governance inheritance is the commercial point: Microsoft's enterprise controls, audit logging, data-residency guarantees, and regulatory posture transfer to the Copilot deployment as a matter of course, and that inheritance is what makes a partner-level risk review survivable without six months of legal negotiation.

The second argument is continuity. From Copilot, the path forward runs through a single architectural family: Copilot Studio for firm-specific bots, Copilot Agents for workflow automation, custom applications on Claude or equivalents for the mature end of the curve. At no point does the firm rebuild, re-platform, or migrate. Each step runs on infrastructure the firm already owns. Each step generates the data foundation the next step requires. For a practice that answers to partners on capital allocation every quarter, that continuity is worth more than a ten-per-cent capability advantage from a competing tool that sits outside the Microsoft perimeter.

The next move

The firms in the PrimeGlobal network that will pull ahead over the next eighteen months are the ones that moved from conversation to deployment inside this budget cycle — starting with a Copilot rollout that compounds into the next build, and the one after that. That's the only call that needs to be made this quarter. The first deployment. Everything else follows.

The question the opening raised is the one that still matters: whether your firm is capturing the time AI has handed back — or still paying people to do it manually.

2026 will not remember who planned to adopt AI. It will only remember those who did.