Claude Now Connects to Spotify, Uber, Instacart, TurboTax and a Dozen More — Every Plan, Today
Anthropic shipped a big piece today: personal-app connectors are live in Claude across every plan — Free, Pro, Max, Team, and Enterprise. The launch roster is heavy: Spotify, Uber, Uber Eats, Instacart, Intuit TurboTax, Intuit Credit Karma, Booking.com, Audible, AllTrails, TripAdvisor, Viator, Resy, StubHub, Taskrabbit, and Thumbtack. Claude can now pull playlists, order food, book a table, find a hike, surface tax-filing context, and tee up rideshares from inside the conversation. Anthropic says personal-connector data is not used to train models, individual apps don’t see conversation context outside their own turns, and any booking or purchase gets an explicit user confirmation step before the action fires.
This is the “Claude goes after ChatGPT and Gemini on their turf” story. ChatGPT and Gemini have been spending a year stitching consumer integrations; Anthropic has been conspicuously absent from that race while it built out the enterprise and developer stories. Today’s launch is the flip: a 15-connector consumer bundle shipped in one day, to every plan. Two things to watch. First, the action-layer design — “checks with you before it buys” is the right default for trust, but it’s also the friction that will get benchmarked against ChatGPT’s Operator and Gemini’s more aggressive agent moves. Second, the revenue shape — consumer connectors tend to come with affiliate economics, and Anthropic just went on the record yesterday doubling down on the “we won’t run ads” stance, so the monetization model for this stack is going to matter.
India’s Finance Minister Convenes Bank CEOs on Mythos — CERT-In and RBI at the Table
Finance Minister Nirmala Sitharaman held a high-level meeting with heads of scheduled commercial banks Thursday to assess emerging cybersecurity risk from Mythos-class models. The Reserve Bank of India and CERT-In, India’s national computer-incident response team, were both present. The framing from the ministry was explicit: Mythos “weaponizing” software vulnerabilities is now the scenario banks have to harden against, and the meeting covered both the defensive-tooling gap and the speed with which regulators need to move. India joins Japan’s FSA (Friday meeting with MUFG, SMFG, Mizuho, BoJ, and the Tokyo Stock Exchange), Australia’s Home Affairs department (direct working relationship with Anthropic), and the Bank of England on the list of national regulators in active Mythos posture.
The more telling piece for Anthropic: the UK bank rollout Reuters reported “within days” on Tuesday is now a week old and Anthropic has not committed to a new timeline. UK officials are still negotiating access parameters while the Bundesbank’s Joachim Nagel keeps publicly arguing Mythos should be universally available to preserve a level playing field. Read the signal: the bank-by-bank, jurisdiction-by-jurisdiction rollout that was the Project Glasswing plan is now running into the sovereign-regulator filter at every stop. Each national supervisor wants to own the access question. Each wants its own working relationship. The rollout is not stuck, but it’s moving at the pace of multilateral diplomacy rather than the pace of a product launch.
Anthropic Publishes the “Ad-Free” Policy — Why Advertising Is Incompatible With a Helpful Claude
Anthropic posted the formal version of its advertising stance this week: Claude products will remain ad-free, and the company argues publicly that advertising incentives are structurally incompatible with a helpful assistant. The published framing is narrower than the shorthand implies — the policy applies to Anthropic’s products specifically, not to developers building on the Claude API, who remain free to ship advertising in their own downstream products. The timing lines up with today’s consumer-connector launch: when you’re about to send users shopping on Instacart and booking hotels on Booking.com through your assistant, the monetization posture is the first question anyone asks.
The implicit business model is clearer now. Anthropic’s consumer story is subscription-plus-connector, not advertising, not affiliate-revenue-share-with-publishers. The revenue story is ARR — which just crossed $30B annualized on the strength of Claude Code and enterprise — and the consumer connectors are distribution and habit-building for a broader subscription base. Whether Anthropic holds that line when ChatGPT goes full-Operator with commerce attribution and Gemini starts paying premium publishers for placement is the longer-term question, but the policy is on the record now. That’s an asset the next time a regulator, competitor, or customer asks.
The Pro-Pricing Hangover — “My Trust in Anthropic’s Transparency Has Been Shaken”
Three days on, the Tuesday pricing incident is the bigger story than Tuesday’s reversal. Recap: the Claude Code support documentation was updated universally to remove Claude Code from the $20 Pro plan. Head of growth Amol Avasare said on X it was “a small test on ~2% of new prosumer signups” and that “existing Pro and Max subscribers aren’t affected.” Simon Willison and others didn’t buy the framing — everyone who checked the pricing grid saw the new version, the Internet Archive captured it, and the support-doc change was global. Anthropic rolled the pages back within hours, but Willison’s read, which is circulating widely in dev Twitter and Hacker News, is that his “trust in Anthropic’s transparency around pricing — a crucial factor in how I understand their products — has been shaken.”
What makes this more than a product-ops flub: the whole Claude Code developer narrative has been built on transparency, public changelogs, and Anthropic treating developers as peers rather than marks. Silently pulling a flagship feature from the $20 plan and then claiming limited scope when the docs said otherwise is exactly the kind of move that the “Anthropic is different” crowd holds up as what Anthropic isn’t supposed to do. The practical advice for teams: lock your model pin, lock your plan choice where your billing allows it, and — if you’re running production workloads — subscribe the account to the API tier rather than relying on Claude Code bundling, which is the durability posture that has held up best through every pricing wobble this year.
Three Short Opus 4.7 Blips This Morning — 3 to 5 Minutes Each, All Resolved
The status page logged three brief windows of elevated errors on Claude Opus 4.7 today, each 3-5 minutes long, all since resolved. The incidents didn’t affect Sonnet or Haiku tiers and didn’t touch the Claude Code CLI routing path. If you saw an Opus-tier 529 or a capacity timeout this morning, that’s the explanation. This is the fourth week running with short-duration Opus 4.7 blips on the board; they fit the pattern Anthropic acknowledged Monday when it conceded demand was causing “inevitable strain” on reliability.
For production workloads, the retry posture that’s holding up best: exponential backoff with jitter, a circuit-break threshold of ~30 seconds, and a fallback to Sonnet 4.5 for anything tier-1. If you’re on Bedrock or Vertex, the cross-region failover story is the one you actually want right now, because the reliability pattern on Anthropic-direct and the hyperscaler-hosted surfaces has diverged measurably in April.
Zilliz Ships claude-context — A Dedicated MCP Server for Codebase Search
Zilliz Tech released claude-context, a Model Context Protocol server designed specifically for Claude Code that indexes an entire codebase as retrievable context for the agent. The pitch: instead of relying on the built-in grep/glob path through Read and Grep, you hand Claude a semantic index over the repo and the agent gets targeted snippets with much less context pressure. The setup takes about five minutes for a medium repo — install the MCP server, point it at your source tree, it builds the index, Claude Code picks it up through ~/.claude.json.
Two things make this interesting. First, it’s a clean test of whether semantic-retrieval MCPs actually beat the file-tools-plus-grep loop on real agent tasks — the answer in initial community benchmarks is “yes on large repos, marginal on small ones.” Second, it slots into the 500K-character MCP output ceiling that shipped on Claude Code 2.1.116 last week, which is the piece that makes big-repo indexing actually useful inside a single session. If you’ve been running into context pressure on repo-wide refactors, this is the first MCP tool since tree-sitter-indexer that’s worth an hour of eval.
NEC Becomes Anthropic’s First Japan-Based Global Partner — The Fine Print From Yesterday
One more beat on yesterday’s NEC announcement, with detail that came out in Tokyo overnight. NEC is now officially Anthropic’s first Japan-based global partner. The scope covers Claude Cowork as the primary surface for NEC employees (not just Claude Code, as the original wire reports implied), and the joint industry-vertical agent work is scoped to financial services, manufacturing, and local government first. BluStellar Scenario is the delivery platform for customer-facing work. The 30,000-seat number is NEC Group-wide and the timeline is “globally, across 2026,” which is meaningful — this is not a ninety-day thing.
The reason to care beyond the headline: the local-government piece is the bit nobody expected. Japanese municipal and prefectural IT modernization has been a dead zone for foreign AI vendors because of data-localization and language constraints. If NEC’s BluStellar-plus-Claude stack wins even a handful of those contracts, that’s a durable distribution advantage Anthropic can’t buy directly and Microsoft and Google haven’t figured out how to crack. Worth watching for tenders over the next two quarters.
Amazon’s $5B-Plus-$20B-Plus-$100B-AWS Deal — What the Trainium Fine Print Actually Says
Monday’s Amazon news has had a week to settle and the silicon details are now clearer. The headline $5B lands now, with up to $20B more tied to commercial milestones. Anthropic commits over $100B to AWS across ten years and gets up to 5 gigawatts of capacity for training and serving Claude. The chip ladder: Trainium2 is coming online in the first half of 2026, nearly 1GW of combined Trainium2 and Trainium3 by year-end, and the contract extends forward to Trainium4 — which doesn’t exist yet. That’s the tell. Amazon is locking in a foundation-model customer for a chip generation that’s still on the roadmap, and Anthropic is locking in compute for a model generation that’s still in research.
Pair with last week’s Alphabet/Broadcom TPU expansion (3.5GW from 2027 on top of the existing $21B TPU order) and Anthropic’s 2027 compute floor is now spread across three silicon stacks: NVIDIA GPUs, AWS Trainium, Google TPUs. That multi-silicon posture is expensive today but is the right hedge against any one supply chain pinching. For builders, the relevance is less about the chips and more about the implied pricing floor — when your vendor has pre-committed hundreds of billions of dollars of compute over ten years, the “models get dramatically cheaper every twelve months” curve is less automatic than 2024 made it look.
Anthropic Fellows Program Applications Close Saturday — If You’re Going to Apply, This Is the Weekend
Quick signal-boost for researchers on the list: the Anthropic Fellows Program application window closes April 26. The program is a paid six-month placement inside Anthropic’s research org, with a focus on alignment, interpretability, and evals. The 2026 cohort is unusual in that it explicitly includes scaling-era security research — the Mythos-adjacent work — which wasn’t a named track in prior years. If you’re doing serious work at that intersection and haven’t applied, this is the weekend.
Worth noting the broader context: Anthropic’s Long-Term Benefit Trust added Vas Narasimhan (former Novartis CEO) to the board on April 14. The board composition now leans more heavily toward pharma-and-biotech governance expertise alongside the existing AI-safety and policy voices — which is consistent with Anthropic positioning frontier models as regulated-industry infrastructure rather than pure software. For fellow-program applicants, that’s a subtle signal about what the org wants to be in three years.
Today Is the Day Anthropic Stopped Being an Enterprise-and-Developer Company
For eighteen months, the Anthropic shape has been legible: frontier models for enterprise, Claude Code for developers, a chat interface that was competitive but deliberately not the main event. The consumer story was ChatGPT’s. The agent-doing-errands story was Gemini’s. Anthropic picked the high-margin lanes and let the competitors fight it out for the mass-market. Today’s consumer-connector launch is the public flip. You do not ship Spotify, Uber, Instacart, Booking.com, TurboTax, Audible, AllTrails, TripAdvisor, Resy, StubHub, and Taskrabbit into every plan — including Free — unless you want Claude to be the assistant people open to book dinner and plan weekends. That’s a different product posture and a different competitive frame.
The reason it matters now: the enterprise and developer stories are mature enough to fund it. ARR is $30B annualized. NEC just put 30,000 engineers on the platform. GIC in Singapore and the APAC rollout are compounding. Mythos is pulling every G7 regulator to the table. The business doesn’t need the consumer play to work for the numbers to look good in Q4. But the distribution question — who owns the assistant the user types to every day — is the trillion-dollar question that cannot be won from the enterprise side. OpenAI has that answer today. Google has an answer through Android and Search. Anthropic has now started the third-horse race for it. The 15 connectors are day one. The real test is how fast the next 50 ship and whether the action-layer feel — the “check with you before buying” friction — evolves into something that feels like an assistant rather than a form-fill wizard. Six months from now we’ll know if today was the pivot or a feature launch.