Microsoft Copilot in 2026: Work IQ and Agent 365

Microsoft’s 2026 update to Copilot adds two main capabilities: Work IQ, which gives Copilot a persistent understanding of how each user works, and Agent 365, which lets Copilot run multi-step business processes without a human at the keyboard. Both are useful for UK SMBs, and both raise security, governance and licensing questions that need a plan.

This guide explains what is changing, what Work IQ and Agent 365 do, the security picture and a sensible action plan. Lanmark is a Microsoft Direct Cloud Solution Provider. For the base product, start with our guide to Microsoft 365 Copilot Business.

What is changing in Microsoft Copilot in 2026

Two themes run through the update. The first is personalisation: Copilot is moving from a chat assistant that responds to single prompts towards a system that learns how each user writes, reasons and prioritises. The second is agency: Copilot is moving from drafting and summarising to running tasks such as scheduling, raising, approving, escalating and routing work between applications and people.

The update is delivered as feature releases inside the Copilot you already pay for, not as a separate SKU. For most UK SMBs already on Microsoft 365 Copilot Business, the new capabilities arrive in the existing licence over the course of 2026. Some advanced agent capabilities will sit behind a separate Agent 365 add-on, where pricing had not been confirmed for UK SMB tenants.

Any firm that ran a Copilot pilot a year ago should plan a refresh.

Copilot Work IQ: personal context that travels with the user

Work IQ is Microsoft’s term for the layer of personal context Copilot maintains for each user: how they write, the projects they work on, the people they collaborate with most, the documents they reference repeatedly and the patterns in their week.

For an FD, that means Copilot eventually understands which board pack lines need the most rigour, which clients drive the cash flow conversation and which suppliers tend to run late. A draft reply or summary is shaped by that context rather than written from a blank slate.

Work IQ is built from data the user already generates inside Microsoft 365: emails, calendar items, documents, Teams conversations and SharePoint activity. The context is held inside the tenant boundary, governed by your existing retention, eDiscovery and information protection policies. The user can inspect what Copilot has learned, edit it and clear it.

The benefit is real. Early Copilot sessions produced generic output until the user learned to load context into each prompt manually. With Work IQ, that context is loaded automatically, and the output quality at prompt one matches what previously took around 20 prompts to reach.

So is the risk. If sensitivity labels and SharePoint permissions are loose, Work IQ will accelerate the wrong kinds of access. A junior user whose access has drifted over years now has a personalised assistant that surfaces sensitive material faster than they could find it manually.

Agent 365: business processes that run themselves

Agent 365 changes Copilot from an assistant into a system that performs work. An agent is a configured workflow inside Copilot that runs a multi-step process on the user’s behalf, handing off to humans only for an approval or a judgement call.

The realistic SMB use cases are smaller than the stage demos:

• New client onboarding: receives a signed engagement letter, creates the SharePoint folder structure, sets permissions, drafts the welcome email for the partner to review, books the kick-off call and raises tasks in the CRM.
• Supplier invoice triage: reads incoming invoices, matches them to purchase orders, flags exceptions for finance and approves the rest within policy limits.
• Monthly compliance check: walks through a checklist, gathers evidence from named sources, drafts the report and routes it for sign-off.

Each agent is built once and runs continuously. You specify the steps, sources, approvers and exception conditions. Microsoft provides the runtime, the connectors to common applications and the auditing and observability layer.

The administrative shift is significant. Agents act on behalf of users, so they need their own identity in Entra ID, their own permission scope and their own audit trail. A firm that has not modernised its identity and access management since its Copilot rollout will struggle to deploy Agent 365 safely.

Security, governance and the SMB risk picture

Agentic Copilot changes the SMB threat model and the regulatory picture. Three issues sit at the top.

Identity sprawl. Each agent is a non-human identity with permissions, and the number of these in a typical SMB tenant will grow quickly through 2026 if agents are deployed without governance. You need an inventory, an owner per agent and a regular review cycle, as you should already have for service accounts.

Data exposure through agent reasoning. An agent with access to multiple data sources can combine them in ways no individual user was authorised to. A client onboarding agent that reads the engagement letter, the conflicts database and the historical client list could surface a conflict previously known only to a partner. That can be the right outcome or a breach of professional confidentiality, depending on who sees the output. Output routing rules and sensitivity labels need to be applied as carefully as access controls.

Regulatory observability. UK SMBs in financial services and legal services are subject to FCA and SRA expectations on the use of AI in client-facing processes, and the EU AI Act applies to firms with EU clients or operations. Agents must be logged, auditable and explainable. Microsoft Purview, Defender for Cloud Apps and the Microsoft 365 audit logs are the tooling, and configuring them correctly is the work. The NCSC guidance on AI security principles is the right framework reference.

Lanmark’s managed detection and response service gives 24x7 SOC monitoring of the Microsoft 365 estate, so unusual agent behaviour, mass data access and anomalous output patterns can be detected and contained.

What UK SMBs should do now

The action plan for the next 90 days has four steps.

1. Governance audit. Get sensitivity labels, SharePoint permissions, Conditional Access policies and identity hygiene into a known-good state before Work IQ accelerates personal context and before agents act on behalf of users.
2. Copilot maturity review. If you rolled out Copilot in 2024, the 2026 features are a refresh, not a free upgrade. Run a fresh user survey, pull usage data from the Microsoft 365 admin centre and recalibrate training and adoption.
3. Small agent pilot. Pick one well-documented, repeatable, low-risk process, such as a meeting room booking workflow, a routine HR onboarding step or an internal status report. Run it for six weeks with a named owner, capture the time saved and use it to write your agent governance policy.
4. Licensing planning. Where Agent 365 sits behind a separate add-on, build the cost-benefit case per agent. Where features arrive in the existing licence, no spend decision is needed, but review whether the current licence mix still fits.

Where Lanmark fits

Copilot 2026 planning sits inside the AI Capability Ladder, Lanmark’s framework for placing Work IQ and Agent 365 in the context of an SMB’s wider AI maturity. Underneath sit the Microsoft 365 platform work and licensing review; around it sits the managed detection and response service.

See our AI consultancy and generative AI consultancy pages, or contact us to plan Copilot 2026 for your firm.