Explaining Microsoft Work IQ, Agent Types and the Road to the Frontier Firm in Microsoft 365 Copilot

Prefer audio?

Who should listen: Microsoft 365 service owners, Copilot and Power Platform admins, and IT leads who need to translate agent marketing terminology into deployment, licensing and governance decisions.

Guest: Alev Tamer and Chris Wheeler — Senior Partner Solution Architect for AI Business Solutions; Senior Copilot Solution Engineer, Microsoft

Alev Tamer and Chris Wheeler, both Copilot specialists at Microsoft, unpack what sits behind the current agent terminology — the spectrum from declarative agents to Copilot Studio and pro-code builds in AI Foundry, the frontier firm maturity model, and how Work IQ grounds responses in organisational context. They also share agents they have built themselves, from a Teams pub quiz bot to autonomous complaint triage in the NHS, and the governance work needed once agent numbers grow.

Many thanks to Jabra, sponsor of this episode.

Key insights

• There are effectively three build routes for agents: declarative no-code agents created inside Copilot, low-code agents in Copilot Studio, and pro-code solutions in AI Foundry. Choosing the wrong tier is the most common early mistake — most business scenarios never need to leave Copilot Studio. ▶ 12:23
• The 'frontier programme' and the 'frontier firm' are two different things and get conflated: the programme is a preview channel for agents that may or may not reach general availability, while the frontier firm is a three-level maturity model running from personal AI assistant, to team-level agents, to agent-operated but human-led organisations. ▶ 17:32
• An NHS complaint-handling workflow was automated end to end: as soon as an email lands in the complaints mailbox, agents triage and route it without human intervention — a good example of where autonomous rather than conversational agents pay off. ▶ 9:20
• Shared agents in Teams meetings behave differently to personal chat — if one person prompts the agent, everyone in the meeting sees the same answer. Chris built a pub quiz agent on this behaviour, which also means it cannot be used covertly for private answers mid-meeting. ▶ 7:49
• Work IQ is more than Graph data — it combines Graph, memory, personalisation and inferencing. Users can explicitly teach Copilot preferences (British English, preferred output formats) and it will apply them across responses. ▶ 22:11
• Agent governance largely falls back on existing Power Platform and Azure controls today: identity, access control, reporting, lifecycle and billing. Agent 365 is Microsoft's answer to agent sprawl once organisations are running hundreds or thousands of agents. ▶ 30:22
• Azure AI Foundry now carries roughly 11,000–12,000 models, and non-Microsoft models such as DeepSeek can be brought in to drive an agent — useful when a specific model outperforms the default for a given task. ▶ 27:49
• Many organisations have missed that the Copilot included tier now reaches the latest models, so some agentic capability is available without every user being on a paid Microsoft 365 Copilot seat. ▶ 20:09
• Agent proliferation is already visible internally at Microsoft — there are agents to brainstorm agents, agents to optimise agent instructions, even agents to write the marketing materials for an agent. Anything built 12 months ago probably has a better option now, so lifecycle review matters. ▶ 16:28

Insights summarised by AI from the episode transcript, reviewed by the Empowering.Cloud team.

Listen: Apple Podcasts · Spotify · Other platforms