AI is starting to bill like a contractor. This fortnight Microsoft rebuilt Copilot around agents that keep working after you log off — and made those agents pay-as-you-go. In Microsoft's own example, a $30 seat turns into a $73 bill once the heavier work starts. So the question for leaders has changed. It's no longer "should we buy AI?" It's "who is watching the meter, and is the work worth it?" Gartner's answer, from a survey of finance chiefs, is that the biggest thing holding teams back now isn't money or tech. It's AI literacy — people knowing how to use it well. In the region, Microsoft committed more than $10bn to the Gulf through 2030, and Saudi Arabia set national standards for what an AI job actually is. Inside: how to work out what an AI task really costs, how to build AI skills without a big training programme, and our new playbook on budgeting for AI that charges by the task.
What's moving in AI worldwide
Your AI agents now come with a meter
Microsoft rebuilt Copilot around agents that keep working after you log off. And those agents are metered.
$30 → $73 — per user per month in Microsoft's own example, once five complex agent tasks are added to everyday use (Microsoft launch materials, reported by VentureBeat, 25 September 2026)
The product news is big, but the billing news is the part to sit with. Until now most companies bought AI the way they buy software: a fixed price per person, per month. Microsoft has now split that in two. The everyday things — chat, help in Office, standard answers — stay inside the seat. The agent work — Autopilot, the Code app builder, the longer Cowork assignments and the newest, most capable models — runs on Copilot Credits and is billed by how much you use.
Microsoft's own worked example shows why that matters. Its $30 seat covers a sample workload of 15 everyday tasks. Add five complex Cowork assignments to 20 everyday tasks and the modelled monthly cost reaches $73 on one model and $69 on another — more than double. These are Microsoft's illustrations, not real customer bills, but the direction is clear: the more real work you hand to agents, the more your AI bill behaves like a contractor's invoice rather than a licence.
Microsoft has put controls around this. Metered services stay switched off until an administrator creates a spending policy, and there are budgets, limits, alerts and per-team model restrictions. Every Autopilot agent carries its own permissions and a full audit log, and people decide how much freedom it gets. As Satya Nadella put it at the launch: "Every agent has to have an identity." Autopilot itself is still in private preview, and Microsoft hasn't published detailed usage rates yet — so now is the time to plan, before the invoices arrive.
Why it matters: When AI charges by the task, someone has to own the meter. The budget conversation moves from "how many licences" to "which work is worth paying an agent to do" — and that's a question for business owners, not just IT.
Source: VentureBeat
The thing holding AI back is now people knowing how to use it
Gartner asked finance chiefs what's stopping AI paying off. The top answer used to be talent. Now it's AI literacy.
9–10 months — typical time for quick-win finance AI tasks — data extraction, payables and receivables, report creation — to deliver their expected value (Gartner survey of 160 senior finance leaders, 24 September 2026)
This is a quiet but important shift. For two years the story was "we can't hire the AI people we need." Gartner's finance survey says that has changed. Agentic coding tools mean you need fewer specialists to build things. What you need now is a whole team that knows how to use what's been built — how to ask, how to check, and when not to trust the answer.
The survey also gives CFOs a useful rule of thumb on timing. The routine work — extracting data, automating accounts payable and receivable, creating reports — generally delivers its expected value within nine to 10 months. The harder, more valuable work — data management, generating insight, forecasting — takes longer to mature. Gartner's warning is not to let the quick wins crowd out the slower ones, because the slower ones are where better decisions and lower risk come from.
Gartner's advice on literacy is refreshingly practical: skip the big classroom programme. Give people real project assignments, a safe sandbox to experiment in, and short on-the-job tasks. Confidence comes from using the tools on real work, not from watching a webinar about them.
Why it matters: If your AI budget has a line for software and none for skills, it's pointed at the wrong problem. The return on AI now depends on how well your people use it — and that's a training decision you can make this quarter.
Source: Gartner
AI in the region: UAE & KSA
Microsoft puts more than $10bn behind AI in the Gulf
Microsoft committed more than $10bn to the UAE, Saudi Arabia, Kuwait and Qatar through 2030.
$10bn+ — committed by Microsoft through 2030 across the UAE, Saudi Arabia, Kuwait and Qatar (Microsoft, announced 23 September 2026, reported by The National)
Announced on 23 September at the UN General Assembly, Microsoft's new Middle East framework commits more than $10bn through 2030. It covers more cloud and AI infrastructure across the region, investment in subsea and land connectivity to speed up internet links, and a new Middle East digital resilience initiative. Microsoft will also place dedicated cyber-security champions in all four countries and expand its AI for Good Labs. President Brad Smith said conflict in the region has "reinforced the connection between digital resilience and digital sovereignty".
Why it matters: More local cloud and AI capacity means more of your AI work can run inside the country — which makes data-residency and continuity questions easier to answer. It also means more demand for people who can build, run and secure it locally.
What to do: Ask your cloud and AI providers two things this quarter: where exactly will our AI workloads run, and what's the plan if a region goes down? Put both answers in your business-continuity plan, not just the IT file.
Source: The National
Saudi Arabia just wrote down what an AI job is
Saudi Arabia launched national standards for AI roles, skills and training. For employers, that's a job-description template from the regulator.
1.56 million — people have been through SDAIA's data and AI training programmes, including 14,495 specialists and experts (SDAIA via Saudi Press Agency, 15 June 2026)
At the UNESCO Global Forum on the Ethics of AI in Riyadh in mid-September, the Saudi Data and AI Authority (SDAIA) launched five frameworks covering schools, universities, training courses and working professionals. The one employers should read is the Data and AI Professional Standards Framework, built with the Ministry of Human Resources and Social Development. It defines the professions, roles and competencies that make up AI work, to line up workforce skills with what the labour market needs. Another sets quality standards for AI training programmes.
Why it matters: When the regulator defines the roles, hiring and training get easier to benchmark. Expect job titles, certifications and Saudization conversations for AI roles to start referring to these standards — so it pays to map your roles to them early.
What to do: If you hire data or AI talent in KSA, get a copy of the professional standards framework and compare it with your current job descriptions. Where your titles or skills lists don't match, fix them now — before candidates and training providers start using the national version as the yardstick.
Source: GCC Business News
AI agents & your workforce
How to work out what an AI task really costs before the bill arrives
Once agents are billed by usage, "is it worth it?" becomes a real question for every workflow. You don't need a finance model to answer it. You need a clear picture of the task, what it costs today, and what it will cost with an agent.
Five steps to price an AI task
- Pick one task and count it. One task, not a department. How many times a month does it happen? First-pass CV screening is a good test: a busy hiring team might screen 400 CVs a month.
- Cost how it's done today. Hours per task times the loaded cost of the people doing it, plus any outside spend like agencies or overtime. That's your baseline.
- Cost the AI version honestly. Usage charges, plus the human time to check the output. Even with a tool like Hyrra.ai, budget time for a recruiter to approve every shortlist.
- Price the mistakes. What does a wrong answer cost — a rework, a delay, a compliance issue? Higher stakes mean more human checking, which changes the maths.
- Run it for 90 days, then decide. Track actual usage and correction rates. Keep it, change it or stop it based on real numbers, not the demo.
If you can't say what a task costs today, you can't tell whether AI is saving you money. Start there.
How to build AI skills in your team without a training programme
Most AI training is a one-off session people forget within a week. Skills stick when people use the tools on their own work, with a safe place to try things and someone to ask.
Five ways to build AI skills that stick
- Start with the work, not the tool. Ask each person to pick one task from their own week to try with AI. Real work makes the learning stick.
- Give people a safe sandbox. An approved space where they can experiment with no customer data and no risk. Fear of breaking something stops more learning than lack of time.
- Teach checking, not just prompting. The most valuable skill is spotting when the answer is wrong. Have people check AI output against a known-good example every week.
- Make it a short project, not a course. A two-week assignment with a real deliverable beats a day in a classroom. Gartner recommends exactly this mix of projects, sandboxes and on-the-job tasks.
- Use a shared skills yardstick. In KSA, SDAIA's new professional standards framework defines AI roles and skills. Use it, or a simple version of it, to see where each person stands.
People learn AI the way they learned spreadsheets: by using it on real work. Give them the time and the room to do that.
The AI Meter: A Leader's Playbook for Budgeting AI That Charges by the Task
Playbook. Six steps for when AI stops being a flat licence and starts billing like a contractor. How to find the meters you already have, price work by the task, give every agent a budget and an owner, set spending guardrails before you switch anything on, measure value honestly, and build the skills that make the spend worthwhile. Written for C-suite and directors in the UAE and Saudi Arabia.
Spotlight: AI-Assisted Hiring with Hyrra.ai
Hyrra.ai. Hyrra.ai puts AI agents to work on the CV pile — it does the first-pass screening and cuts the time it takes to build a shortlist, with a recruiter in control of every shortlist and every rejection. It's exactly the kind of task this edition says to price up first: high volume, clear rules, and a person who owns the decision.