In the span of about four weeks, enterprise software went from “AI is coming” to “AI is here, it has integrations, it’s running for months unattended, and it’s coming for your SaaS stack.” February 2026 wasn’t a single announcement — it was a cluster of product moves from Anthropic and Perplexity that, taken together, describe a fundamentally different relationship between knowledge workers and software. Not AI as a chatbot you consult. AI as a coworker who sits in your tools, understands your files, and executes multi-step work without you babysitting each step.
Here’s what actually happened, why it matters, and what you should be thinking about right now.
Anthropic Stopped Playing Defense
For most of 2024 and into 2025, Anthropic’s public positioning was careful, safety-first, and slightly hesitant to push hard into enterprise. That changed fast. By the end of January 2026, they had shipped Claude Cowork — a desktop app (macOS first) that runs Claude in an isolated VM on your local machine, with full access to local files and MCP integrations. By February 17, Claude Sonnet 4.6 was live, at the same price as 4.5 but with meaningfully better performance, a 1M token context window in beta, and improved agentic search that uses fewer tokens to do more.
Scott White, Anthropic’s Head of Product Enterprise, described the Cowork vision as “transitioning into vibe working” — knowledge workers directing AI rather than writing code themselves. That framing is deliberate. It’s the enterprise equivalent of what “vibe coding” did for developers: lower the floor on who can use powerful AI capabilities without technical overhead.
The numbers behind this aren’t marketing. Anthropic’s own engineers reportedly use Claude for roughly 60% of their work and are shipping 60 to 100 internal releases per day. Claude Cowork itself was built using Claude Code in 10 days. That’s not a case study — that’s a live demonstration of the product eating its own cooking at scale.
Meanwhile, Anthropic deprecated Opus 4 and 4.1 from the model selector, pushing users toward Opus 4.6 (now the flagship, with a 1M token context window and improved coding) and Sonnet 4.6 for everyday tasks. They also launched self-serve Enterprise plans — no sales call required — and announced a $100M Partner Network on March 12. The company is reportedly approaching $19 billion in annualized revenue. This is not a research lab hedging its bets. This is a company that has decided to compete.
Claude Code Became an Actual Platform
If you haven’t been watching Claude Code closely, you’ve missed something significant. What started as a developer-facing CLI tool has been receiving daily releases and is quietly becoming a serious software development platform in its own right.
The Skills API — organized folders with SKILL.md files — lets you create modular, reusable agent behaviors. Pre-built skills now cover PPTX, XLSX, DOCX, and PDF generation. The --channels permission relay (research preview) and --bare flag for scripted automation suggest Anthropic is building toward agentic pipelines that can be embedded deeply into existing developer workflows, not just used interactively.
Claude Code is now included in every Team plan standard seat. That’s not a small detail. It means any organization on a Team plan can start experimenting with agentic coding workflows without an additional line item. Combined with 1M context for Opus 4.6 by default on Max, Team, and Enterprise plans, plus voice mode and improved MCP tool support, the product is closing the gap between “AI assistant” and “AI that ships things.”
Anthropic’s Frontier Red Team used Claude Opus 4.6 to find more than 500 vulnerabilities in production open-source code. That’s a meaningful proof point — not because it’s surprising that a capable model can do security research, but because running that kind of systematic analysis across production codebases at scale was previously the domain of expensive consultancies and dedicated security teams.
Perplexity’s “Computer” Is Not a Computer — And That’s the Point
On February 25, Aravind Srinivas launched what Perplexity called Perplexity Computer. The name is a little misleading — it’s not hardware. It’s a cloud-based multi-agent system, described as “a general-purpose digital worker.” The framing of “computer” is intentional: Perplexity is positioning this not as a software product you use, but as a worker you delegate to.
The architecture is interesting. Rather than building a single powerful model, Perplexity Computer orchestrates 19 to 20 frontier models, routing tasks to whichever model does them best. Claude Opus 4.6 handles core reasoning. Gemini handles deep research and spawning sub-agents. Grok handles speed on lightweight tasks. Veo 3.1 handles video. GPT-5.2 handles long-context recall and wide search. The system is explicitly designed to be model-agnostic — as better models emerge, they get swapped in.
The internal benchmark Perplexity shared is striking: the system reportedly saved $1.6 million in labor costs, performing 3.25 years of work in four weeks across 16,000 queries. Those numbers should be read with appropriate skepticism — self-reported benchmarks from companies with valuation pressure deserve scrutiny — but even directionally, they describe a different category of productivity tool than anything available 18 months ago.
The system can run autonomously for “hours or even months.” It connects to 400+ app integrations. Available to Perplexity Max subscribers at $200 per month.
Then on March 11, at Perplexity’s Ask 2026 developer conference, they went further: Personal Computer runs on a dedicated Mac mini (M4, max RAM), running 24/7 and connected to both local apps and Perplexity’s cloud. Computer for Enterprise adds Slack integration (just @computer in any channel), Snowflake, Salesforce, and HubSpot connectors, SOC 2 Type II compliance, SAML SSO, 40+ financial data tools including SEC filings and FactSet, and an AI-native browser for organizations called Comet Enterprise.
The competitive positioning is explicit: Perplexity is going after Microsoft Copilot and Salesforce. The company is valued at $20 billion, was at roughly $148M ARR in mid-2025, and is targeting $656M by end of 2026. That’s an aggressive ramp, and it depends entirely on enterprise adoption going faster than anyone expected.
What These Two Moves Have in Common
The surface-level story is “two AI companies launched new products.” The deeper story is convergence on the same architecture: persistent agents, local + cloud hybrid execution, deep integration into existing tools, and human oversight that’s opt-in rather than mandatory at every step.
Both Claude Cowork and Perplexity Computer assume that the bottleneck is no longer AI capability — it’s integration and trust. Anthropic solved for trust by running in an