September decides your IT year. While budgets reopen and steering committees make their calls, the CIO who shows up with a clear roadmap gets ahead of the one who’ll “look into it after the summer.” Here are the five AI workstreams to lock down before December 2026 — the practical synthesis of a summer of news, each item with a direct bearing on your Q4 decisions.

Why fall 2026 is a turning point for CIOs

Summer 2026 was eventful: the Microsoft 365 price increase on July 1, the partial postponement of the AI Act just days before the August 2 deadline, the merger of the Copilot apps, and mounting pressure from finance leaders to prove AI’s return on investment. These signals point to one reality: the experimentation phase is over, and disciplined execution begins. According to Foundry’s AI Priorities 2026 study, 65% of enterprises increased their AI budget this year, with a median rise of 22% — but CFOs now demand clear metrics before renewing the investment.

The plan below is not a technology wish list. It’s a five-workstream roadmap, ordered by urgency, that I recommend to my CIO and SMB-leader clients for the September–December quarter. Each links to a detailed article published this summer.

Workstream 1 — Budget and licensing: decide before your renewals

The Microsoft 365 price increase that took effect on July 1, 2026 is no footnote: Microsoft 365 E3 rose from $36 to $39 per user per month, E5 from $57 to $60, and the frontline F3 plan from $8 to $10. A crucial and often-missed detail: existing customers keep their current price until their first renewal after July 1. In other words, your renewal date is now a budget variable in its own right for Q4.

On the Copilot side, the standalone license stays at $30 per user per month, while Copilot Business for SMBs starts at $18 (a promotional annual rate through September 2026) before moving to $21. The window to lock in the promo rate closes during your back-to-school quarter. Before you sign, ask the one question that matters: how many of your Copilot licenses are actually used each week? If you don’t know, don’t renew blind — audit first. I laid out the decision grid by company profile in my breakdown of the July price increase.

Workstream 2 — AI Act compliance: what actually applies since August 2

This is probably the most misunderstood point of the season. The Digital Omnibus on AI — Regulation (EU) 2026/1744, published in the Official Journal on July 24 and in force since July 27, 2026 — did postpone some deadlines, but not all. Obligations for stand-alone high-risk Annex III systems (HR, credit, education…) move to December 2, 2027, and those for AI embedded in regulated products (Annex I) to August 2, 2028.

But beware the misreading: other obligations do apply, as of August 2, 2026. The Article 50 transparency requirements (telling users they’re talking to a chatbot, labeling deepfakes and generated content) remain in force. And a new prohibition was added to Article 5 regarding AI-generated non-consensual intimate imagery. For a Copilot user, that means, concretely: map your high-risk AI uses, document user-facing transparency, and above all don’t loosen your governance on the pretext of a “postponement.” The topic deserves dedicated ownership, as I explained in my article on AI governance in 2026.

Workstream 3 — Copilot adoption: relaunch before the curve collapses

The figures released this summer are unambiguous. Microsoft closed fiscal 2026 with 30 million Copilot seats, and customers with more than 50,000 seats grew sevenfold year over year. Yet fewer than 4.5% of the 450 million commercial Microsoft 365 users pay for Copilot — and among those who do, only 20 to 30% use it weekly. Real penetration hovers around 6.5%.

That gap between licenses purchased and actual usage is the number-one destroyer of enterprise AI value. I described from the field why 70% of Copilot rollouts stall after three months and the associated relaunch plan. For Q4, the concrete action fits in one sentence: stop counting licenses, count weekly active usage, and relaunch by business use case rather than by generic communication. HR, for instance, offers immediate wins — I shared 15 Copilot prompts for September’s hiring campaigns.

Workstream 4 — AI agents: frame before you deploy

68% of CIOs rank AI agents among their top three investment priorities for 2026, and Microsoft introduced a new tier of background AI agents in August. The appetite is real. So is the risk: Gartner predicts 40% of agentic AI projects will be canceled by 2027, mainly those with weak business cases or insufficient data foundations.

The back-to-school lesson isn’t to give up on agents, but to refuse to deploy them without a quantified use case and a clean data foundation. An agent that automates a poorly defined process merely industrializes the mess. Pick one or two high-repetition, measurable processes, and instrument them before scaling. It’s also the right moment to settle which model fits which task, which I covered in my comparison of ChatGPT, Claude and Copilot.

Workstream 5 — Governance and data: the non-negotiable foundation

The previous four workstreams all rest on the same base. Analysts agree: the 2026 priority is shifting toward strong data foundations, modern integration and clear governance. In practice, buyers now value audit trails, access controls and the ability to oversee or override an agent’s behavior. The advisory consensus is clear: the CIO must establish a formal AI governance framework in partnership with legal and risk, aligned with the AI Act.

Without a designated owner of AI governance, everything else stays fragile. It’s the foundation on which leadership trust is built — and the condition for AI budgets to be renewed in 2027.

Your Q4 checklist at a glance

In short, here are the five decisions to make before December: audit real license usage before any Microsoft 365 renewal; map your high-risk AI uses and document the transparency the AI Act requires; shift from license-based tracking to weekly active-usage tracking; deploy agents only on quantified use cases built on clean data; and appoint an owner of AI governance. Five points, one quarter, an IT year that’s decided now.

Let’s build this roadmap together

Every organization decides differently depending on its size, sector and maturity. If you want to turn this checklist into a concrete plan tailored to your context — license audit, AI governance, Copilot adoption relaunch — let’s talk on a discovery call. I support SMBs and mid-market companies as a fractional CIO, precisely on these topics.

📕 Going further — Check out my two books, packed with concrete use cases and prompts to leverage AI day to day: IA Générative pour tous and Microsoft Copilot pour les Freelances (available in paperback and ebook on Amazon).

Article written by Sylvain Jacquemard — AI & Digital Transformation Expert | sylvainjacquemard.blog

Sources

• Foundry, AI Priorities Study 2026; Jade Global, 2026 AI Trends for CIOs; Gartner (agentic forecast, via StackAI / MarketScale).
• TeamAscend, iFeeltech, TTMS — Microsoft 365 Pricing Updates 2026.
• getpanto.ai, Stackmatix, TechTimes, empowering.cloud — Microsoft Copilot adoption statistics, August 2026.
• Gibson Dunn, Cloud Security Alliance, Center for Cyber Diplomacy — EU AI Act Digital Omnibus / What August 2 Means, Regulation (EU) 2026/1744.