Something changed in how AI gets sold this fortnight. Salesforce released seven AI agents with first names and job descriptions — Casey on customer help, Hunter on outbound sales, Paige on IT and HR. Not features. Roles. One of them can now chase a goal for weeks instead of one conversation. So the thing you are buying looks less like software and more like a hire. Which raises an awkward question: if the agent got a job description, who rewrote the job descriptions of the people around it? McKinsey's new survey says the answer is mostly nobody, and that the big companies are pulling away — 40% of firms over $1bn are scaling agents now, against 22% of smaller ones. In the region, the UAE did the people side properly: it approved an AI curriculum for every school and put 22,000 teachers through training to teach it. And in Riyadh, HUMAIN started raising outside money to pay for the compute underneath all of this. Inside: how to write a job description for an agent, how to spot which of your roles changes most, and our new playbook on redrawing the org chart.
Salesforce just released seven AI agents with first names and job titles. You are no longer buying a tool. You are filling a role.
7 billion — Agentic Work Units delivered across Agentforce and Slack in two years — 3.2 billion of them in Q2 2026 alone (Salesforce, 11 September 2026)
The detail that matters here isn't the names. It's that each agent arrives with a defined scope, a set of actions and a place in a process — the same things you write down when you hire someone. Companies can even rename them to fit their own brand, which tells you exactly how Salesforce expects them to be treated.
The long-horizon runtime is the other shift. Until now an agent did a task and stopped. Hunter can take an objective like "rescue my at-risk deals this quarter", turn it into a plan, work at it over weeks, and come back to a human at the points where it needs a decision. Three things make that work: memory that survives between sessions, execution that keeps running and can course-correct, and guardrails that define when it acts alone and when it waits.
Salesforce published customer numbers alongside the launch — 50% of Engine's chat enquiries resolved by its help agent, 60% of Perk's sales pipeline built by its outbound agent, 70% of Autism Queensland's admin requests handled by its employee agent. These are vendor-reported figures, not audited benchmarks, so read them as direction of travel. The direction is clear enough: agents are being measured on the share of a job they complete, not on how clever they sound.
Why it matters: If the thing you are buying has a job description, it belongs in your workforce planning, not just your software budget. That means a named manager, a clear scope, a way to measure the work, and a decision about what the people around it now do instead.
Source: Salesforce Newsroom
In one year, large firms went from 27% to 40% on scaling AI agents. Smaller firms didn't move at all.
40% vs 22% — of large companies (over $1bn revenue) are scaling AI agents, up from 27% — against 22% at smaller companies, unchanged (McKinsey, The state of AI in 2026, 25 August 2026)
There are two useful numbers buried in this survey and neither is the headline.
The first is the size gap. Large employers added 13 points of agent scaling in twelve months while smaller ones added nothing. That is not a technology gap — the tools are the same price for everyone. It's a gap in who has someone whose actual job is to make this work.
The second is what separates the small group getting real financial results from everybody else. Nearly three-quarters of the high performers say they have fundamentally redesigned workflows because of AI, up from 55% last year. Among everyone else, it's one quarter. They are also twice as likely to say their senior leaders visibly back the work, and twice as likely to actually measure its impact. The pattern is consistent: value shows up when the work is redesigned, not when a tool is bolted onto the work you already had.
One more thing worth holding on to. 39% of respondents expect AI to reduce headcount over the next year. But only 14% reported an actual reduction over the past year — against the 32% who predicted one in last year's survey. Expectations have run well ahead of reality two years running. Plan for the work changing, not for the people disappearing.
Why it matters: The build-versus-buy line has moved. A third of organisations have already cancelled a software purchase because coding agents made building it cheaper, which changes what skills you need on the payroll. And if you're a mid-sized Gulf employer, the competitors above you just pulled a year ahead on agents.
Source: McKinsey & Company
The UAE approved an AI curriculum for every school in the country, and funded the retraining of 22,000 teachers in the same decision.
22,000 — teachers and educators to be trained to teach and use AI, alongside a new AI curriculum for every public and private school (UAE Cabinet via WAM, 2 September 2026)
At the Cabinet meeting on 2 September, chaired by Sheikh Mohammed bin Rashid, the UAE approved an AI curriculum for all public and private schools. It covers the technical side of AI plus ethics, data, algorithms, applications and risks. The stated goal is to train 22,000 teachers and educators to use AI in teaching, assessment, curriculum analysis and lesson planning. The Cabinet also approved a way to offer the curriculum to education bodies in other countries. The same meeting unveiled the Cabinet AI Advisor — 32 specialised AI assistants that analyse policies and legislation, assess their impact, check global practice and recommend to ministers.
Why it matters: Two things follow for employers. The graduates you hire from here on will have been taught AI at school, so your baseline talent is about to improve without you lifting a finger. And the government just modelled the step most companies skip: it didn't only approve the technology, it funded the retraining of the people who have to use it, in the same paper, with a number attached.
What to do: Copy the shape, not the scale. Take the one team whose work changes most when an agent arrives, write down the skills they'll need, and put a budget line and a date against training them — in the same document that approves the tool.
Source: The National
HUMAIN is preparing to list and raising outside capital. The kingdom's AI build-out is moving from state cheque to real balance sheet.
68MW → 467MW — Saudi data-centre capacity between 2021 and Q1 2026, on more than SR56.2bn ($14.98bn) of investment in data centres and digital infrastructure (Official Saudi data, reported by Fortune, 9 September 2026)
HUMAIN's chief executive Tareq Amin has started assembling a team to prepare an IPO, with a stated aim of listing in both Saudi Arabia and New York by 2029. Bloomberg reported the company is also seeking a $2.5bn fund for a new wave of data-centre expansion, and Amin said the $10bn global AI venture fund announced last year could launch larger than planned by the end of 2026. That fund will only back companies that commit to using Saudi compute or building a workforce in the kingdom. HUMAIN is targeting 1.9GW of AI compute by 2030 and more than 6GW by 2034.
Why it matters: "Take the money, but put your compute or your people in the kingdom" is industrial policy wearing a venture-capital jacket. If you hire in Saudi Arabia, that's a direct signal about where technical roles, data-centre operations and AI delivery work will sit over the next five years. It also means the sovereign-AI conversation in procurement is getting more specific than "where does our data live".
What to do: If you operate in KSA, ask your cloud and AI vendors two questions this quarter: which kingdom region will our workloads actually run in, and what local hiring or training commitment comes with the contract. Both are becoming standard asks.
Source: Fortune
When an agent shows up pre-named and pre-scoped, the temptation is to just turn it on. Then three months later nobody can say who owns it, what it's allowed to do, or whether it's working. Fifteen minutes of writing at the start prevents all of that.
If you wouldn't hire someone without a job description, don't deploy an agent without one.
Most AI training budgets get spread thinly across everyone, which helps nobody very much. The better move is to find the handful of roles where the work is genuinely changing, and spend the money there.
You can't train for a change you haven't located.
Playbook. Six steps for redrawing the org chart when part of the work moves to software. What the agent owns, who manages it, how the human job descriptions change in the same week, and how to keep building junior talent when the routine work that used to train them disappears. Written for C-suite and directors in the UAE and Saudi Arabia.
Hyrra.ai. Hyrra.ai puts AI agents to work on the CV pile — it does the first-pass screening and cuts time-to-shortlist, with a recruiter in control of every shortlist and every rejection.