Code with Claude 2026: 5 lessons for UK SMEs from the London keynote

Code with Claude 2026 visited London on 19 May, sat between the San Francisco kickoff on 6 May and the Tokyo close on 10 June, and the announcements that landed there set the agenda for every UK small and medium business deciding how to use Claude in the next twelve months. Anthropic did not ship a new model on stage. What it did ship matters more for a UK accountancy firm, recruitment consultancy or marketing agency than another benchmark tick: managed agents, multi-agent orchestration, doubled rate limits, and a clear signal that the next year of competitive lift for Anthropic comes from how Claude works, not how much Claude knows.

Why Code with Claude 2026 mattered for a UK audience

The London leg of the conference is the part most UK firms missed in the noise. Anthropic chose London as its second of three stops, ahead of Tokyo, which is a clear statement that the UK market sits among Anthropic’s top three commercial priorities. Senior Anthropic leaders, including Brad Abrams, ran the London sessions on managed agents and the new Claude Code SDK in person, and a number of UK enterprise customers including major UK retail banks were in the room. The reason that matters for a UK SME is the work being done on agent governance in London directly answers the questions UK regulators are now asking of every AI deployment.

The headline number Anthropic CEO Dario Amodei dropped on stage is the one a UK CFO should read. “First-quarter 2026 revenue and usage, on an annualised basis, grew 80x rather than the 10x Anthropic had planned,” Amodei said. For a UK SME deciding whether to commit to Claude over Microsoft Copilot or Google Gemini for the next renewal cycle, that growth number is two things at once: a vote of confidence in the platform’s staying power, and a warning that capacity may be tight if every UK firm makes the same decision at the same time. The doubled rate limits Anthropic shipped at the keynote are a direct response to that capacity pressure.

Managed Agents: what changes for a UK SME this quarter

Claude Managed Agents is the announcement with the most immediate UK SME impact. The product gives a UK business the ability to deploy a Claude-powered agent that runs sandboxed code on Anthropic’s infrastructure, with checkpointing so a long-running task can pause and resume, and credential scoping so the agent only sees the data it needs to do the job at hand. For a UK accountancy firm that wants Claude to reconcile a client’s Xero ledger against a HMRC submission, the credential scoping means the agent gets the Xero API token for one client at a time, and nothing else. That is the answer to the Information Commissioner’s Office “minimum necessary data” principle.

Multiagent Orchestration is the second piece, and it is the one that changes how a small UK team can scale. Until now, a UK SME using Claude has been writing prompts for a single agent at a time. Multiagent Orchestration lets a single prompt break into a fleet of specialised agents that hand work between each other. For a UK marketing agency producing campaign content, that means one agent can draft copy, another can fact-check it against the client’s brief, a third can generate the supporting social posts, and a fourth can prepare the brief for the human reviewer, all without a marketing manager scheduling the steps. The keynote demo did this for a marketing campaign in under three minutes.

The Outcomes feature: success criteria UK clients will actually read

Outcomes is the third keynote feature, and it is the one a UK SME owner should be reading the documentation on right now. The idea is to let a user define what success looks like for a given Claude task, in plain English, and have Claude iterate until it meets those criteria before returning a result. For a UK solicitor’s office asking Claude to draft a Section 25 notice under the Landlord and Tenant Act 1954, the success criteria could be: “cites the correct sections, identifies the relevant ground from a-g, includes the required statutory wording, returns a result the partner does not need to redraft.” Outcomes lets Claude self-check against that list before handing back a draft.

The practical effect for a UK SME is a shift in what work looks like. A junior solicitor today drafts a Section 25 notice, a partner reviews it, sends it back with corrections, the junior redrafts, the partner reviews again. With Outcomes, the partner defines the success criteria once, the agent iterates against them, and the partner reviews the result. The first iteration is much closer to acceptable, which compresses the review loop. UK firms that already use Claude.ai or the API should be testing Outcomes against their highest-volume drafting workflows this month, ahead of any procurement decisions in the autumn.

Dreaming and what session memory means for UK confidentiality

Dreaming is the most novel and most carefully framed announcement Anthropic made. The feature lets Claude recall what happened in previous sessions and build on past work, which is the persistent memory that every previous Claude generation explicitly did not have. The framing matters because UK confidentiality rules and the Information Commissioner’s Office’s expectations around data minimisation cut directly against “AI that remembers everything.” Anthropic addressed that by making Dreaming opt-in per workspace, with explicit user controls on what gets remembered, and with credential-scoping that mirrors the managed agents pattern.

For a UK SME, the practical guidance is simple. Turn Dreaming on for internal-only workflows where remembering matters, such as an account manager building a long-running brief on a single client. Leave Dreaming off for any workflow where the data leaving Anthropic’s infrastructure would breach a contractual confidentiality obligation, particularly if you handle UK financial services, legal services, or healthcare data. The default Anthropic recommends, which is opt-in per workspace rather than account-wide, is the right one for any UK firm to adopt as policy.

The new pricing reality: usage-based and what it means for UK budgets

Anthropic shifted from flat-fee enterprise pricing to a usage-based model at Code with Claude 2026, which is the change a UK finance director should be paying the most attention to. The old structure billed a fixed monthly fee per seat with capped usage; the new structure bills per token consumed, with discounts for higher commit volumes. For a UK SME that uses Claude heavily for a small subset of users, this is good news: usage scales with actual work done, not with seat count. For a UK SME that bought a fleet of Claude licences and uses them lightly, the cost may rise once the renewal lands.

The Vercel data point Anthropic shared on stage is the one to internalise. “Opus tokens represent around 23% of usage but more than 70% of spend.” Translated to a UK SME budget, that means three quarters of a Claude bill will sit on the Opus tier of model, even if the majority of prompts use Sonnet. Most UK SMEs will need to set a per-team policy on when an Opus prompt is justified, because the cost difference is real and visible in the new pricing. Brad Abrams at Anthropic put the trade-off plainly: “We get close to Opus level intelligence at much lower prices because we are being very conservative about the tokens.” UK firms should expect to set Sonnet as default and Opus as deliberate opt-in.

Claude Code, the developer story, and UK contractors

Claude Code crossed USD1bn in annualised revenue in May 2026, which makes it the fastest-growing developer tool in Anthropic’s history and a serious competitor to GitHub Copilot for UK independent developers and small studios. The Code with Claude keynote added remote session control across devices, a split-view GUI, an Auto Mode that classifies permissions automatically, and routines for scheduled execution via cron or webhooks. For a UK contractor billing by the hour to a UK client, the routines feature alone changes the maths. A scheduled Claude Code routine can run nightly maintenance, refactor, and test passes against a client’s codebase without consuming the contractor’s daytime hours, which can either lower the bill to the client or free hours for new business.

The Mario Rodriguez quote from the keynote stage is the one any UK developer should hold onto. As GitHub’s Chief Product Officer, Rodriguez emphasised that cache hit rates above 94% are now foundational, and that “1% efficiency means millions overall.” For a UK independent developer running Claude Code locally against a UK client’s codebase, what Rodriguez was saying is that Anthropic and GitHub have figured out how to make the per-prompt cost of an AI development workflow drop sharply as a project’s history accumulates. The first week on a new codebase is the expensive week; the months that follow are dramatically cheaper. That changes the contractor pricing conversation.

Claude vs Copilot vs Gemini: the UK SME decision tree

For a UK SME buying once and renewing for a year, the practical choice still comes down to Claude, Microsoft 365 Copilot, or Google Gemini Workspace. The Code with Claude announcements sharpen the case for Claude in three specific UK use cases. First, any UK SME where confidentiality is a contractual or regulatory obligation should prefer Claude for the constitutional AI safety framing that has held up under Information Commissioner’s Office scrutiny better than the competing systems. Second, any UK SME running multi-step content or analysis work benefits more from Multiagent Orchestration than from Copilot’s chat-and-edit pattern. Third, any UK SME with serious developer headcount should prefer Claude Code, which beats GitHub Copilot on routines and remote sessions.

Microsoft 365 Copilot still wins when the SME is already deep on Microsoft 365 with Word, Excel, Outlook and Teams as the primary tools, because Copilot lives inside those apps natively. Google Gemini Workspace wins when the SME runs on Workspace and uses Gmail, Docs and Sheets as the primary tools. Claude wins when the SME wants to use AI as a layer above any productivity stack, with explicit per-workflow controls and credential scoping. For most UK SMEs, the answer is to run two: Claude for analysis and Copilot or Gemini for in-app productivity, and the new usage-based Claude pricing makes that combination cheaper than it was three months ago.

The honest summary for UK SME buyers

If you run a UK SME today, the practical to-do list out of Code with Claude 2026 is short. Turn on Multiagent Orchestration for your highest-volume content or analysis workflow this month and measure the time saved. Set Outcomes against the three drafting tasks your team redoes most often. Leave Dreaming off until you have written a workspace policy for it that your data protection lead has signed off. Move budget planning from flat-fee assumptions to usage-based, and budget for an Opus default cost overshoot of around 70% on the Sonnet baseline.

And if you are a UK developer or run a small UK dev studio, the more important question is whether Claude Code’s routines and remote session control change what your billable day looks like. The answer for most UK contractors is yes, and the firms that adapt first will be the firms that win UK enterprise work over the next twelve months. For our take on how Anthropic’s roadmap fits UK regulated industries specifically, see our recent guides on Claude for UK accountants and the upcoming Claude for UK solicitors safe-use guide, both of which apply directly to the Outcomes and Dreaming pattern Anthropic shipped at the keynote.

What we like, what we’d watch

Sources

• InfoQ, “Anthropic’s Code with Claude Announces Managed Agents, Proactive Workflows, Capability Curve”, May 2026.
• Anthropic, “Code with Claude 2026: Opening Keynote”, YouTube, 6 May 2026.
• Simon Willison, “Live blog: Code w/ Claude 2026”, 6 May 2026.
• MIT Technology Review, “Anthropic’s Code with Claude showed off coding’s future—whether you like it or not”, 21 May 2026.
• Anthropic Claude Pro UK pricing
• Anthropic UK enterprise contact
• HMRC IR35 contractor guidance
• ICO guidance on AI and personal data
• AWS UK regions data residency
• Stripe UK pricing

UK reader FAQ

Related on MTW

• Best AI writing assistant UK 2026
• Microsoft Build 2026 UK keynote
• Microsoft Copilot at Accenture UK lessons

MMTW Editorial