AI Leadership Weekly · Issue #70 · Tuesday 10 February 2026 · 08:00 GMT

Good morning.

OpenAI introduced Frontier as a platform for enterprise agents. Anthropic released Opus 4.6, and GitHub opened access to additional coding agents in public preview. The announcements bring the management of AI work closer to the tools people already use. For a business buyer, the immediate question is which capabilities are genuinely available to your team and what remains an early customer programme.

OpenAI introduces Frontier. OpenAI announced an enterprise platform for developing and managing agents, initially working with a limited group of customers.

Opus 4.6 is released. Anthropic launched Opus 4.6, including a one-million-token context window in beta and updated capabilities for longer assignments.

More agents come to GitHub. GitHub opened a public preview of Claude and Codex coding agents for eligible Copilot users within its existing development environment.

https://openai.com/index/introducing-openai-frontier/

OpenAI · 5 February 2026

What happened. OpenAI introduced Frontier on 5 February as a platform for building, deploying and managing agents across business systems. Its announcement emphasised shared context, evaluation, permissions and identity, with initial availability for a limited set of customers. It described a platform direction and early deployments, not universal access or guaranteed customer outcomes.

Why it matters. Our take: The management layer deserves attention because someone must own the task after the demonstration ends. Ask how an agent is introduced, supervised, changed and retired. Keep responsibility attached to the underlying service. A dashboard that shows activity is useful only if the organisation knows who should respond when that activity is wrong.

What to do. Write a one-page role description for one proposed agent: purpose, inputs, actions, output, owner and escalation. Ask the supplier to show how those decisions are represented in its platform. Test one failed assignment and one permission refusal, not just a successful happy-path demonstration.

https://www.anthropic.com/news/claude-opus-4-6

Anthropic · 5 February 2026

What happened. Anthropic released Claude Opus 4.6 on 5 February, highlighting coding, longer tasks and professional work such as analysis and document creation. The release included a one-million-token context window in beta and additional controls for effort. The supplier’s capability comparisons do not remove the need to check outputs or define a safe operating scope.

Why it matters. Our take: More capacity to handle context is not permission to connect every record. Select the information necessary for a task and make the output traceable. For a complex assignment, decide what should be reviewed along the way rather than waiting for a large final artefact whose mistakes are difficult to untangle.

What to do. Take a difficult but bounded analysis task and specify the evidence pack and expected answer. Ask for an assumptions list and unresolved questions. Review the output against the sources. Compare the entire effort with the current approach before widening the brief or adding more access.

https://github.blog/changelog/2026-02-04-claude-and-codex-are-now-available-in-public-preview-on-github/

GitHub · 4 February 2026

What happened. GitHub announced on 4 February that Claude and Codex were available in public preview alongside its existing agent capabilities. The rollout had subscription and administrative conditions. Bringing additional agents into GitHub does not mean that repository rules, review responsibilities or the need to control access disappear.

Why it matters. Our take: Treat the coding agent as a contributor within your delivery process. The useful question is whether it can produce a change that another person can understand, test and maintain. Keep the same quality and release expectations across suppliers so the evaluation does not reward whichever tool was given the easiest assignment.

What to do. Choose a small set of comparable maintenance tasks and a common completion checklist. Require changes to be explained and tested before review. Keep credentials and production actions outside the trial. Ask maintainers whether the submitted work is easier to own, not merely whether it arrived quickly.

20 January · ChatGPT adds age prediction. OpenAI described age prediction for consumer accounts, with additional protections for suspected under-18 users and a route to correct mistakes. Source

16 January · OpenAI outlines advertising tests. OpenAI announced plans to test advertising for eligible adult US Free and Go users; the announcement preceded the test itself. Source

15 January · Anthropic examines how AI is used. The January Economic Index analysed sampled Claude interactions using measures such as task complexity and autonomy; it was not a workforce census. Source

14 January · Gemini adds Personal Intelligence. Google began an opt-in US beta connecting personal Google information. The initial offer did not extend to Workspace business accounts. Source

More dated updates from the preceding 30 days.

2 February · Codex gets a desktop app. OpenAI launched a macOS app for managing parallel coding-agent tasks. Windows availability was not part of the initial release. Source

27 January · Prism puts AI inside scientific writing. OpenAI launched Prism, a free research-writing workspace powered by GPT-5.2 for personal accounts; organisational-plan access was still forthcoming. Source

26 January · Microsoft introduces Maia 200. Microsoft unveiled an accelerator designed for AI inference, with initial deployment in its US Central datacentre region. Source

22 January · Anthropic publishes Claude’s constitution. Anthropic set out the principles intended to shape Claude’s behaviour. The document describes training intentions rather than guaranteed outcomes. Source

Minutes 0–4 · Define the assignment. Name the business process and the person receiving the result. Describe the work the agent is intended to perform. Add a short list of activities it must not perform, even if technically capable of doing them.

Minutes 4–8 · Specify authority. Separate access to information from permission to make changes. Record approval points, spend or retry limits and the circumstances that require escalation. Make the boundaries visible to both the user and the person supporting the service.

Minutes 8–12 · Set supervision. Choose the evidence needed to accept the output and the checks required during longer work. Name the reviewer and the person responsible for failed tasks. Avoid a model owner being mistaken for the business owner of every workflow.

Minutes 12–15 · Plan retirement. Describe how to stop new work, preserve the audit record and reassign unfinished tasks. Confirm that permissions can be removed. Set a review date for the role card before expanding the agent’s scope or changing its underlying model.

Design the responsibility around the agent before scaling the technology. A clear role, a useful acceptance test and a reliable handover make it easier to improve components without losing ownership of the service.

Who owns the outcome when an agent crosses from one application into another?

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

Explore Data Wave