Anthropic Publishes Labor Market Impact Study — AI Exposure High, Actual Displacement Low
Anthropic released a comprehensive research paper measuring how AI interacts with the workforce using actual Claude usage data rather than theoretical capability assessments. The study covers 800+ occupations and introduces a new metric that weights fully automated tasks more heavily than human-assisted ones. The headline finding: computer programmers have 75% of their tasks technically exposed to AI, followed by customer service representatives and data entry keyers at 67%. Financial analysts, technical writers, and legal researchers also rank high.
But here's the critical nuance: despite high theoretical exposure, the study finds no measurable impact on unemployment rates for workers in the most-exposed occupations. There is tentative evidence that hiring has slowed slightly for 22–25 year-olds entering those fields. The demographic profile of the most-exposed worker skews higher-paid and better-educated — 17.4% hold graduate degrees versus 4.5% in the unexposed group. The gap between what AI can technically do and what companies are actually deploying remains enormous.
API Spring Cleaning — Sonnet 3.7 and Haiku 3.5 Retired, New Features Go Live
Anthropic has retired Claude Sonnet 3.7 (claude-3-7-sonnet-20250219) and Claude Haiku 3.5 (claude-3-5-haiku-20241022) — all API requests to these models now return errors. Claude Haiku 3 (claude-3-haiku-20240307) has been formally deprecated with retirement set for April 19, 2026. Developers still on these models need to migrate to Sonnet 4.6 and Haiku 4.5 respectively.
On the feature side, several capabilities have graduated: structured outputs are now GA on Sonnet 4.5, Opus 4.5, and Haiku 4.5 (with output_format moved to output_config.format). Fine-grained tool streaming is GA across all models with no beta header required. The compaction API is in open beta on Opus 4.6, enabling server-side context summarization for effectively infinite conversations. Data residency controls let developers specify US-only inference at 1.1x pricing. And a new auto-caching feature eliminates manual cache breakpoint management with a single cache_control field.
Compaction API Opens for Long-Running Agentic Workflows
The compaction API (beta header: compact-2026-01-12) is now available on Opus 4.6 and solves one of the biggest practical constraints in agentic development: context window exhaustion during long-running tasks. When enabled, Claude automatically detects when input tokens exceed your configured threshold, generates a summary of the conversation so far, creates a compaction block, and continues with the compressed context. No manual truncation logic required.
This is particularly significant for developers building autonomous agents that need to maintain coherent state across hundreds of tool calls. Previously, you'd either hit the context ceiling and crash, or build custom summarization middleware. Now it's a single API parameter. The feature is available on the direct API, AWS Bedrock, and Azure — check the docs for platform-specific beta headers.
Enterprise Analytics API and Self-Serve Enterprise Plans Now Available
Two enterprise-focused updates shipped this week. The Enterprise Analytics API now provides programmatic access to usage and engagement data for Claude and Claude Code Remote usage within organizations, aggregated per org per day. This gives engineering and finance teams the data they need to track adoption, allocate costs, and justify spend without scraping dashboards.
Separately, enterprise plans are now available for self-serve purchase directly on the Anthropic website, with a single seat type covering Claude, Claude Code, and Cowork. Previously, enterprise access required going through a sales team. This removes a friction point for smaller companies that want enterprise features but don't want to negotiate a contract.
Tom's Guide Tests Claude Inside Slack, Figma, and Asana — Hands-On Verdict
Published today, Tom's Guide ran Claude's MCP-powered integrations through real workplace scenarios. In Figma, Claude generates multiple CTA button variants with reasoning behind each option and adjusts tone to match brand voice — keeping designers in creative flow instead of bouncing between tools. In Slack, Claude reads specific channels and threads to help teams stay on top of discussions without manual catch-up. In Asana, the integration targets the bottlenecks and status drift that accumulate in project management tools over time.
The underlying pattern is significant: Anthropic isn't building separate apps for each platform. Claude connects to these tools via MCP from within the Claude chat interface, so your entire work context lives in one place. This is the "workplace command center" thesis that VentureBeat flagged when these integrations first launched in January — and the Tom's Guide review suggests it's working in practice, not just in demos.
Anthropic IPO Watch — $380B Valuation, Law Firm Hired, Race with OpenAI
Anthropic closed a $30 billion funding round in February 2026 at a $380 billion valuation — and the IPO machinery is visibly in motion. The company has hired Wilson Sonsini to prepare for a public offering, and has had preliminary talks with investment banks. An Anthropic spokesperson told Reuters the company has not decided when or if it will go public, but multiple investors are actively encouraging Anthropic to file ahead of OpenAI. Betting platform Kalshi gives Anthropic a 72% chance of going public before OpenAI does.
The Pentagon controversy has been a double-edged sword: it creates political risk, but it also drove Claude downloads up 240% month-over-month in February and pushed the app to #1 on Apple's App Store. Annualized revenue is projected to nearly triple to around $26 billion. For enterprise customers evaluating long-term Claude commitments, the IPO timeline matters — a public Anthropic would have different incentive structures, disclosure requirements, and potentially different pricing strategies.