AI Leadership Weekly · Issue #74 · Tuesday 10 March 2026 · 08:00 GMT

Good morning.

OpenAI’s GPT-5.4 release and Microsoft’s Copilot Cowork preview both point towards AI taking on more of an assignment. Anthropic’s labour-market research asks a related but different question: what does observed AI use tell us about jobs? The useful distinction is between a product’s stated capability, what people actually use it for and evidence of a wider economic effect.

GPT-5.4 adds computer use. OpenAI launched GPT-5.4 across ChatGPT, the API and Codex, with native computer-use capabilities through developer and coding routes.

Microsoft previews Copilot Cowork. Microsoft announced Copilot Cowork for longer, multi-step work, initially in limited research preview with broader Frontier access planned.

Anthropic studies labour-market exposure. Anthropic published an early analysis of AI exposure and labour-market outcomes, distinguishing observed usage from theoretical task coverage.

https://openai.com/index/introducing-gpt-5-4/

OpenAI · 5 March 2026

What happened. OpenAI launched GPT-5.4 on 5 March across ChatGPT, the API and Codex, describing improvements in professional work, coding and tool use. It introduced native computer-use capabilities in the API and Codex. Those are capabilities of a released model and its access routes, not evidence that an entire business role can be handed over without redesign.

Why it matters. Our take: Break an assignment into preparation, analysis, decision and communication. A tool may contribute differently to each. For example, you might test it on assembling an evidence pack while keeping a commercial judgement with the accountable person. Make that division explicit, so a useful contribution does not become an unexamined transfer of authority.

What to do. Ask a team to choose one recurring assignment and mark the steps that are repetitive, judgement-heavy or dependent on relationships. Test a bounded contribution with copies of approved material. Evaluate the complete handover to the next person, not just the quality of a document produced in isolation.

https://www.microsoft.com/en-us/copilot/blog/2026/03/09/copilot-cowork-a-new-way-of-getting-work-done/

Microsoft · 9 March 2026

What happened. On 9 March, Microsoft announced Copilot Cowork for work across Microsoft 365, developed with Anthropic technology. The original announcement described a limited research preview, with broader access through its Frontier programme expected later in March. It was not general availability for all Microsoft 365 customers at the time of this edition.

Why it matters. Our take: A cross-application assignment can conceal several separate permissions: reading a document, editing a file, arranging work and communicating externally. Decide which are needed for the initial service instead of granting the whole chain because the demonstration is convenient. Ask users how they would recognise a wrong assumption before it affects someone else’s work.

What to do. Write a realistic brief for a cross-application task and list each system it would touch. Mark the steps that may prepare a draft and those that require approval before changing or communicating anything. Check your organisation’s actual preview access before treating the task as an available team service.

https://www.anthropic.com/research/labor-market-impacts

Anthropic · 5 March 2026

What happened. Anthropic published a US labour-market study on 5 March combining a measure of observed AI exposure with employment data. It reported no systematic increase in unemployment for highly exposed workers in its analysis, while identifying a tentative hiring signal for younger workers. The findings were early evidence with limitations, not proof of causation or a forecast for UK employers.

Why it matters. Our take: Do not translate an occupational exposure measure directly into a local headcount target. Examine which tasks your team performs and how the work reaches an acceptable outcome. Ask where junior colleagues learn judgement today, and what would replace that experience if preparation work changed. A redesign should make development deliberate rather than leave it to chance.

What to do. Bring a manager and two people doing the work into a short discussion. Compare the proposed AI contribution with the actual mix of tasks and learning opportunities. Record what the evidence supports, what remains uncertain and how the team would test the change without pre-deciding a staffing outcome.

19 February · Gemini 3.1 Pro enters preview. Google introduced Gemini 3.1 Pro in preview, targeting more complex reasoning tasks across its consumer and developer products. Source

17 February · Sonnet 4.6 arrives. Anthropic released Sonnet 4.6, reporting improvements in coding, computer use and reasoning while retaining the Sonnet pricing level. Source

12 February · Codex-Spark targets fast coding. OpenAI introduced GPT-5.3-Codex-Spark as a research preview for Pro users, initially focused on text-based, low-latency coding. Source

12 February · Gemini Deep Think expands. Google offered its updated reasoning mode to AI Ultra subscribers and invited interest in early API access, not general API availability. Source

More dated updates from the preceding 30 days.

27 February · Amazon and OpenAI expand their relationship. The companies announced a strategic partnership including investment, compute and a planned stateful runtime on Amazon Bedrock. Source

26 February · Nano Banana 2 launches. Google introduced Nano Banana 2, combining image-generation capabilities with a greater emphasis on speed and broader product access. Source

24 February · Cowork gains enterprise plugins. Anthropic introduced organisation-specific plugin distribution and customisation, allowing teams to package connected tools and reusable skills. Source

20 February · Claude previews code-security review. Anthropic announced a limited research preview of Claude Code Security, designed to identify vulnerabilities and propose fixes for human review. Source

Minutes 0–4 · Describe the real week. List five recurring activities with someone who performs the role. Include coordination, exceptions and learning, not just formal outputs. Identify which activities consume effort and which carry responsibility for a consequential decision.

Minutes 4–8 · Propose a bounded change. Choose one activity where assistance could make the work easier. State what the person retains, what the tool prepares and what the next recipient receives. Avoid using a whole job title as the scope of the experiment.

Minutes 8–12 · Protect the learning. Ask how less-experienced colleagues currently develop judgement through this activity. Decide how they would practise and receive feedback after the change. Include time for that learning in the proposed process instead of treating it as overhead to eliminate.

Minutes 12–15 · Agree the evidence. Select a quality measure, a workload measure and a feedback point with the team. Run a limited trial and discuss the result together. Keep the decision about expanding the change separate from assumptions about future staffing.

Make AI adoption a conversation about better work, supported by visible evidence. Put the business outcome, the people doing the work and the responsibility for decisions in the same design—not in three disconnected plans.

Which specific activity should change, and how will the people doing it help define a better outcome?

Brought to you by Data Wave — your AI & Data Team as a Subscription.

Explore Data Wave