AI in February 2026: the agent platforms arrive
February was one of the busiest months the AI industry has seen. Three of the biggest developers shipped new models within a fortnight, OpenAI launched a platform for running agents across a business, and investors started pricing in what agents could mean for traditional software. Here is what happened and why it matters.
A fortnight of new frontier models
On 5 February, Anthropic released Claude Opus 4.6 and OpenAI released GPT-5.3-Codex on the same day. Opus 4.6 brought a one-million-token context window in beta, the ability to run teams of agents in parallel in Claude Code, and improved work in Excel and PowerPoint, with pricing unchanged. OpenAI described GPT-5.3-Codex as its most capable agentic coding model to date, noting it is 25% faster than its predecessor and stronger at general computer tasks.
On 17 February, Anthropic followed with Claude Sonnet 4.6, a mid-priced model it says developers in its early testing preferred to its previous top-tier Opus 4.5 about 59% of the time, with notable gains in operating software such as spreadsheets and web forms. Two days later, Google began rolling out Gemini 3.1 Pro in preview, claiming double the reasoning performance of Gemini 3 Pro.
What it means for your business: The capability you could only get from the most expensive model a few months ago is now available at mid-tier prices. If you evaluated AI for a process last year and found it not quite good enough, it is worth testing again. Avoid locking yourself into a single provider: design your systems so you can switch models as prices and performance change.
OpenAI launches Frontier to manage agents at scale
Also on 5 February, OpenAI introduced Frontier, an enterprise platform for building, deploying and managing AI agents. It connects agents to a company's data warehouses, CRMs and internal tools, gives each agent its own identity with explicit permissions, and includes built-in evaluation to improve performance over time. OpenAI named HP, Intuit, Oracle, State Farm, Thermo Fisher and Uber among the first adopters, with broader availability promised over the following months.
What it means for your business: The conversation is shifting from "which chatbot should we use?" to "how do we run and govern many agents safely?" Identity, permissions and monitoring for agents are becoming as important as they are for human staff. Even if you are not ready for an enterprise platform, start treating each automation or agent as something that needs an owner, defined access rights and a way to measure its results.
Agent plugins rattle software and data stocks
At the end of January, Anthropic released plugins for its Cowork agent aimed at legal, sales, marketing and data analysis work. In early February, Reuters reported a sharp sell-off in shares of companies seen as exposed. Thomson Reuters fell nearly 18%, its largest daily loss on record, while RELX dropped 14%, Wolters Kluwer about 13% and LegalZoom 19.7%. One Schroders analyst said investors were "aggressively repricing" these areas as AI threatened traditional software licensing models.
What it means for your business: Share prices are not a verdict on any product, but the reaction shows how seriously markets now take the idea that agents could replace parts of specialist software and outsourced services. When your software contracts come up for renewal, ask what an agent-based approach could do, and whether your vendors have credible AI plans of their own.
EU AI Act high-risk guidance slips
The European Commission missed its 2 February deadline to publish guidance on how to classify high-risk AI systems under Article 6 of the AI Act, IAPP reported. The Commission said it was incorporating months of feedback and expected to publish a draft later in the month. At the same time, its proposed Digital Omnibus package, which would push back high-risk obligations currently due in August 2026, was still working its way through the EU institutions.
What it means for your business: If you sell into the EU or use AI in areas such as recruitment, credit or essential services, the rules still apply, but the detail and timing remain uncertain. Do not wait for final guidance to act: keep an inventory of where you use AI, record what each system does, and note which uses could fall into high-risk categories.
Practical takeaways
Re-test any AI use case you parked last year; mid-priced models have improved significantly.
Build automations so the underlying model can be swapped without rebuilding the whole process.
Give every agent or automation a named owner, clear permissions and simple performance measures.
Review upcoming software renewals with agent-based alternatives in mind.
Start or update an AI inventory now so you are ready when EU AI Act guidance is finalised.
If this month's pace has left you unsure where to focus, you are not alone. We would be happy to help you make sense of the options and find the opportunities that fit your organisation. Book a call with Manta Global AI to explore what is possible.
Comments