Code with Claude London Closes With “Extended” — A Full Day for Indie Devs and Founders, Not Enterprise Buyers
Anthropic added a second London day this year aimed entirely at independent developers and early-stage founders, and it runs today. Where Wednesday was the keynote-and-breakouts format streamed to a global audience, “Extended London” is the in-room, laptops-open counterpart: founder stories, builder deep-dives, and hands-on workshops led by Anthropic’s Applied AI team, with demos and office hours running all day.
The split is deliberate. The 2026 tour has leaned hard into agent infrastructure — the platform, the tooling, the connectors — rather than a headline model drop, and an Extended day is how Anthropic feeds that back to the long tail of solo builders who actually wire the MCP connectors and ship the apps. If you couldn’t get into the keynote, this is the track designed for you; expect the practical, “here’s how to build it” material to surface from the workshops over the next few days.
SAP Names Claude Its Primary Reasoning Engine — Anthropic’s Models Get Embedded Across the SAP Business AI Platform
At SAP Sapphire, SAP and Anthropic announced an expanded collaboration that makes Claude a primary reasoning and agentic capability embedded across SAP’s AI-enabled solution portfolio, including the newly announced SAP Business AI Platform. In plain terms: Claude becomes a default brain for agentic workflows running on top of the systems where enterprises already keep their finance, supply-chain, procurement, and HR data.
This is the kind of distribution that’s hard to overstate. SAP sits inside a large share of the world’s biggest companies, and embedding Claude at the platform layer — rather than as a bolt-on chat window — means agents can reason directly over the system of record. It lands the same week as the London platform keynote and the KPMG and PwC alliances, reinforcing the pattern: Anthropic keeps monetizing the model through deep partner integration into enterprise plumbing.
Cache Diagnostics Hits Public Beta — the API Now Tells You Exactly Why a Prompt-Cache Prefix Missed
Prompt caching is one of the biggest cost levers on the Claude API, and it’s also one of the easiest to break silently — a single byte of drift in your prefix and the cache quietly misses. The new cache diagnostics beta fixes the guesswork: pass diagnostics.previous_message_id on a Messages request and the API returns a cache_miss_reason explaining where the current prompt diverged from the previous turn. Add the cache-diagnosis-2026-04-07 beta header to opt in.
If you run long-context agents or RAG pipelines, this is a meaningful quality-of-life win — cache misses translate directly into latency and dollars, and until now you mostly debugged them by trial and error. Wire it into your eval harness, catch the drift, and you can claw back the savings prompt caching is supposed to deliver. It’s a small feature with an outsized payoff for anyone watching their token bill.
Fast Mode Comes to Opus 4.7 — Up to 2.5x Faster Output Tokens From the Same Model, at a Premium
Fast mode — still a research-preview beta — now supports Claude Opus 4.7. Set speed: "fast" with model: "claude-opus-4-7" and the fast-mode-2026-02-01 beta header to get significantly faster output token generation: up to 2.5x higher tokens per second from the same model, at premium pricing. Pricing, rate limits, and access mirror the existing Opus 4.6 fast mode, and interested customers join via waitlist.
The use case is narrow but real: latency-sensitive, Opus-quality work — live coding assistants, interactive agents, anything where a user is watching the tokens stream. You’re paying for speed, not smarts, so it’s a knob to reach for only when responsiveness is the bottleneck. For batch or background jobs, standard Opus 4.7 is still the cheaper call.
Claude Code Sharpens Background-Session and Plugin Management — /resume for Background Agents, agents --json, and Pre-Install Plugin Inspection
The Claude Code changelog kept its weekly cadence. Background sessions started via claude --bg or Agent view now show up under /resume alongside interactive ones (marked bg), and background subagent completion notifications include elapsed duration. For scripting, claude agents --json lists live sessions as JSON — handy for status bars, tmux-resurrect, or custom session pickers — and OTEL spans gained agent_id and parent_agent_id attributes with fixed trace parenting.
On the ergonomics side: the /plugin Discover and Browse screens now show a plugin’s commands, agents, skills, hooks, and MCP/LSP servers before you install it, so there’s no more blind installs. /model now changes the model for the current session only (press d in the picker to set a new default), and “extra usage” was renamed “usage credits.” Plus the usual reliability sweep — fixes for startup hangs when the API is unreachable, garbled output after window resizes, and image-resize bugs.
Anthropic Backs a $1.5B AI-Native Enterprise Services Firm With Blackstone, Hellman & Friedman, and Goldman Sachs
Anthropic is part of a $1.5 billion joint venture to stand up a new AI-native enterprise services company, alongside Blackstone, Hellman & Friedman, and Goldman Sachs. Anthropic, Blackstone, and Hellman & Friedman are each putting in roughly $300 million, with Goldman Sachs contributing about $150 million. The venture is built to deliver Claude-powered services to large organizations.
Read it next to the KPMG, PwC, and SAP deals and the strategy is unmistakable: rather than only selling tokens, Anthropic is investing in the delivery layer — the consultancies and services firms that actually implement AI inside the Fortune 500. Owning a slice of that channel both accelerates deployment and captures more of the value Claude creates downstream. It’s a capital-intensive bet on distribution, not just on the model.
The $200M Gates Foundation Partnership Keeps Surfacing — Claude Aimed at Global Health, Education, and Smallholder Farming
Anthropic’s four-year, $200 million partnership with the Gates Foundation continues to draw attention as more details land. The commitment — grant funding, Claude usage credits, and technical support — targets global health, life sciences, education, and economic mobility, with the largest share aimed at health outcomes in low- and middle-income countries, starting with overlooked diseases like polio, HPV, and preeclampsia.
The education and economic-mobility threads are worth watching for the public goods they’ll produce: evidence-based K-12 tutoring tools in the US, foundational literacy and numeracy apps in sub-Saharan Africa and India, and agriculture-specific model improvements plus local-crop datasets and benchmarks. The first public benchmarks and datasets for AI math tutoring are slated for release later in 2026 — useful artifacts even if you’re nowhere near the development sector.
Thursday Read — The Same Week, Two Audiences: SAP at the Top of the Funnel, Extended London at the Bottom
Today’s pairing is a neat snapshot of how Anthropic is playing the whole board at once. At the top: SAP makes Claude the primary reasoning engine across its Business AI Platform, and a $1.5B services JV lines up the consultancies to deploy it inside the Fortune 500. At the bottom: the Extended London day puts Anthropic’s Applied AI team in a room with solo founders, teaching them to wire connectors and ship apps. Same week, opposite ends of the funnel — and both feed the same flywheel.
The connective tissue is the developer platform, which is exactly where today’s quieter news lives. Cache diagnostics and fast mode for Opus 4.7 aren’t headline-grabbers, but they’re the kind of unglamorous improvements that make Claude cheaper and faster to build on at scale — the substrate under both the SAP integration and the indie app. The throughline of this whole tour holds: the model number matters less every week, and the durable advantage is the platform, the tooling, and the connector graph teams have already wired up. If you build on Claude, today’s homework is short — turn on cache diagnostics, see where your prefix is drifting, and reclaim the caching savings you’re probably leaving on the table.