Claude 4 API: Tool Use Streaming & Prompt Caching vs Cursor 1.5

“The primitive here is clean: incremental tool-call deltas over SSE, and a cache-control header you attach to prompt segments to pin them server-side. The DX bet is that complexity lives in the HTTP layer, not in a new SDK abstraction — you opt in per-request, no new mental model required. The moment of truth is calling `stream=true` on a tool-use request and watching partial JSON arguments arrive before the model finishes thinking, which actually matters for agent loops where you want to dispatch work early. This is not a weekend-script replacement — implementing correct incremental JSON parsing for partial tool arguments plus a reliable distributed cache with 2M token capacity is a real engineering problem Anthropic has solved for you. The specific decision that earns the ship: cache invalidation is explicit and cache hits are reflected in the usage object, so you can actually measure what you're saving instead of guessing.”

“The primitive here is asynchronous agent execution decoupled from IDE focus — finally, you can kick off a refactor or test-writing task and context-switch without the whole thing dying. The DX bet is correct: the complexity is hidden in the runtime, not pushed onto the developer via config or orchestration boilerplate. The moment of truth is queuing a multi-file task, closing the tab, and coming back to a diff — and apparently it survives that test. Shared team rules is the feature that actually earns the enterprise tier: replacing the tribal knowledge of per-developer .cursorrules files with a versioned, shared config is the kind of mundane-but-real problem that unlocks actual team adoption. The autocomplete latency improvement is the only claim I'd want benchmarks on before citing it.”

“Direct competitors are OpenAI's cached completions and Google's context caching in Gemini 1.5 — both shipping for months — so Anthropic is catching up, not leading. The specific scenario where this breaks: cache hit rates depend entirely on prompt structure, and developers who dynamically compose system prompts (inserting user-specific context at the top) will see near-zero cache utilization and pay full price while assuming they're saving money. The prediction: this feature doesn't get killed — it becomes table stakes infrastructure and Anthropic wins by having the largest cache window (2M vs. competitors' current limits). What would have to be true for me to be wrong: OpenAI ships a 10M token cache window before Anthropic's ecosystem matures, commoditizing the advantage. Still a ship because the streaming tool-use delta is genuinely differentiated — no competitor has clean partial-argument streaming for tool calls yet, and that changes agent loop architecture in ways that matter.”

“Background agent execution is the one feature that separates Cursor from GitHub Copilot in a meaningful, non-cosmetic way — Copilot hasn't shipped async task delegation at the IDE level, and that gap is real enough to matter today. The scenario where this breaks is multi-repo or monorepo tasks that cross service boundaries: background agents operating on partial context without a human in the loop will produce confident wrong diffs, and the memory panel won't save you there. What kills this in 12 months isn't a competitor — it's OpenAI or Anthropic shipping native IDE integrations with the same async primitive baked into their own tooling, collapsing the moat. But right now, the team rules feature alone justifies the Business tier for any eng team above 10 people, so this ships.”

“The thesis this bets on: by 2027, the dominant AI application architecture is a persistent agent with a large, stable context (tools, memory, instructions) that gets reused across thousands of user interactions — making context I/O cost the primary unit economics lever, not generation cost. The dependency that has to hold: agents don't collapse back to stateless chatbots, and context windows keep growing faster than per-token prices fall. The second-order effect nobody's talking about: prompt caching at 2M tokens makes it economically viable to give every enterprise user a fully-loaded, role-specific agent context at request time — which shifts competitive differentiation from 'who has the best model' to 'who has the best cached context corpus,' effectively making knowledge curation the new moat. This tool is riding the trend of context-window expansion-as-infrastructure, and it's on-time, not early — but the streaming tool-use primitive is ahead of the curve on agent loop efficiency. The future state where this is infrastructure: every production agentic system has a cache manifest the same way it has a CDN config.”

“The thesis Cursor 1.5 is betting on: within two years, developers will manage fleets of concurrent async coding tasks rather than typing code themselves, and the IDE becomes a task dispatcher rather than a text editor. Background agent execution is the first real infrastructure bet on that trajectory — not a demo, an actual runtime change. The dependency that has to hold is that agents remain good enough to be trusted with multi-step tasks but not so good that the IDE layer becomes irrelevant entirely; Cursor is threading a specific needle in that window. The second-order effect nobody is talking about: shared team rules start to function as organizational AI policy, meaning the eng team — not IT, not legal — becomes the de facto owner of how AI behaves in the codebase. That's a power shift worth watching. Cursor is early on the async-agent trend line and building the right primitives for it.”

“The buyer is the engineering team at any company running Claude in production with long system prompts or multi-step agents — this comes out of the AI infrastructure budget, not a new budget line, which means no procurement friction. The pricing architecture is sound: cache reads at ~90% discount means the savings are real and measurable in the first billing cycle, which creates immediate retention — developers who restructure prompts to maximize cache hits are now architecturally coupled to Anthropic's caching implementation. The moat question is the honest one: this is infrastructure that OpenAI and Google will match, so the defensible position isn't the feature itself but the ecosystem of developers who've restructured their codebases around it. What survives a 10x model price drop: the streaming tool-use architecture, because that's about latency, not cost. The specific business decision that makes this viable is pricing cache reads as a separate SKU — it lets Anthropic capture value from high-volume production workloads without losing price-sensitive experimenters.”

“The buyer here is clear: VP Eng or CTO at a 20-200 person company, paid from the dev tooling budget, justified by reduced context-switching cost and standardized AI behavior across the team. Shared team rules is the expansion revenue mechanism — it's the feature that converts individual Pro subscribers into Business accounts, and that's a real land-and-expand wedge built into the product itself rather than bolted on by a sales team. The moat question is harder: Anysphere's defensibility depends on workflow lock-in through memory and rules accumulation, which gets stickier the longer a team uses it, but the underlying model access is still commoditized. The risk is that VS Code's own AI layer catches up fast enough that the switching cost never fully sets. For now, the unit economics on the Business tier are credible.”