The move: We can see their move to corner the cloud-native parallel agent layer — Cursor 2.0 shifts the primary surface from the editor to coordinated cloud environments, with Bugbot as the first autonomous agent closing the loop on PRs. The JetBrains ACP integration is equally significant: Cursor is no longer a VS Code fork but an agent platform that runs where developers already are. They are now competing with GitHub's Coding Agent as directly as they compete with Windsurf.

The move: We can see their two-track move: at the model layer, Claude leads on safety, long context, and reasoning; at the platform layer, Claude Code is becoming a programmable agent OS — hooks let teams intercept every lifecycle event, subagents enable parallel task trees, and skills make workflows reusable. Donating MCP to a neutral foundation removes the "Anthropic-controlled standard" objection and accelerates ecosystem adoption — so the same context layer serves Claude, Cursor, Xcode, and ChatGPT alike.

The move: We can see their breadth play: Codex CLI for developers who want local-first tools; Codex Enterprise for teams; Operator for browser/computer automation; ChatGPT for consumers. The Astral acquisition is the most strategically interesting move — buying ruff and uv means OpenAI is inserting itself into the developer toolchain at the linter and package manager layer, not just at the model layer. Together, this is a bet that the AI company that touches the most points in the developer workflow wins.

The move: We can see their move to corner the "codebase as context, CI as execution" layer — the Coding Agent closes the loop from issue triage to merged PR entirely within GitHub infrastructure. The Copilot SDK means teams can build custom agents on top of Copilot's proven foundation, not just consume it. For organisations that standardise on GitHub, this makes Copilot the natural agent platform — no separate tool required.

The move: Donating MCP to a neutral foundation was the move that turned a promising protocol into a genuine standard. With Apple Xcode, OpenAI ChatGPT, Cursor, and JetBrains all adopting MCP natively, the "Anthropic-controlled standard" objection is gone. The connector layer is now effectively settled: teams and vendors building MCP servers can reach every major AI client without re-implementing integrations per product.

The move: We can see their two-track move on coding: build Jules as a native cloud agent (plan → clone → execute → PR, all on Google infrastructure) while simultaneously licensing Windsurf's IDE UX innovations and hiring its founding team. This means Google is buying its way into the editor experience layer while building the agent layer from scratch. With Gemini 3 Pro's advanced thinking now powering both, the "grounded, enterprise, autonomous" position is the most complete Google has ever had in developer AI.

The move: We can see their move to corner productised autonomous execution with the Manus acquisition — a task-execution agent with persistent runs and a work-app connector set, while keeping an open layer (OpenManus). The speed of the deal (~10 days) and the commitment to wind down China operations signal that Meta sees autonomous task agents as strategically critical, and that national security considerations are now a real constraint on AI M&A. Manus's revenue trajectory ($125M+ ARR in 8 months) also validates demand for general-purpose autonomous agents beyond coding.

The move: We can see their move to corner the "incumbent IDE + agentic AI" layer — 2026.1 ships AI Agents and Agent Skills so developers who live in JetBrains get multi-step agentic task execution without switching tools. The ACP dynamic is notable: by accepting Cursor agent traffic, JetBrains became a distribution surface for the broader agent ecosystem, not just a consumer of Anthropic or OpenAI models.

The move: Windsurf is the defining consolidation story of 2025–26. Microsoft's veto of the OpenAI acquisition shows that big-tech partnership agreements now constrain M&A at scale — a warning for any AI startup reliant on a single cloud partner. The asset split — technology to Google, product to Cognition — means Windsurf's innovations feed two different directions simultaneously. Watch Cognition (Devin + Windsurf IDE) as the combined autonomous-agent-plus-editor play, and watch how Google's licensing of Windsurf UX shapes Gemini Code Assist in 2026.

The move: We can see their move to corner the "all-in-one builder cloud" layer — Agent 4's parallel execution means Replit is not just a coding environment but a concurrent development platform where multiple specialised agents build different parts of the stack at the same time. Adding mobile closes the last gap in their "build anything, ship anywhere" story. Turbo pricing shows confidence in monetising agent speed as the primary value driver.

The move: We can see their move to corner the enterprise and cloud AI layer with a "managed everything" position: Q Developer for the IDE, Bedrock for model choice (now including OpenAI o3 alongside Claude and Llama), and AgentCore for the agent runtime. The OpenAI partnership is the most notable shift — it removes the single-model dependency concern that has kept some enterprise teams off Bedrock, and signals that AWS is betting on model plurality rather than model exclusivity.

The move: Firecrawl is cornering the "web context as a service" layer — the piece that sits between the open web and any AI agent that needs real-world data. Becoming an official Claude MCP server means any Claude Code or Cursor workflow can reach the web via a single tool call. As agents are increasingly asked to research, monitor, and act on live information, Firecrawl is the path of least resistance for that data layer.

The move: GStack is the first serious open-source attempt to answer the question: what happens when you engineer the *process* around Claude Code, not just individual prompts? The bet is that the bottleneck to AI-driven software delivery is not model capability but the absence of structure — and that encoding engineering discipline into slash commands unlocks throughput that ad-hoc prompting cannot. Garry Tan's 600K LOC in 60 days is the proof of concept; the open-source release is the invitation for the broader community to iterate on it.

The move: Paperclip is betting that the next bottleneck in AI-assisted development is not individual agent capability but coordination — the same problem that makes human organisations hard. By bringing org structure (roles, delegation, context inheritance) to multi-agent workflows, Paperclip goes beyond "run multiple agents" to "run an AI company with consistent context and governance." The model-agnostic, MCP-first design means it can sit above Claude Code, Cursor, or any other agent runtime — a coordination layer, not a replacement.