In a two‑week sprint this August, the Claude team made the core user experience of claude.ai and the Claude desktop app roughly three times faster. The effort focused on the four journeys that account for about 95 % of user activity: launching the app, starting a conversation, loading an existing conversation, and sending a message. At the 75th percentile, the time to reach a typeable page on a fresh load of claude.ai dropped from 3.1 seconds to 0.55 seconds. Starting a new Claude Code session fell from 0.8 seconds to 0.3 seconds, and loading a Claude Cowork cloud session improved from 2.6 seconds to 0.73 seconds. The team estimates that these gains save tens of thousands of user‑hours of waiting every day. The sprint relied on a dedicated Slack channel where Claude (running an internal research model comparable to Opus 5.5) was tasked with monitoring performance, maintaining observability dashboards, identifying bottlenecks, and proposing improvements. By asking Claude to analyze usage data through the Datadog MCP server, the team identified the four highest‑impact journeys, which translated into thirteen distinct measurements. Baselines were established by adding instrumentation so each metric began with a user interaction and ended once the result was rendered, separating client and server work. The team began with a list of about twenty hand‑picked projects targeting specific journeys. Claude estimated the impact of each in milliseconds, allowing the group to set sprint targets. Twelve of the thirteen targets were met by day three, prompting the setting of new goals as additional opportunities emerged. Among the shipped changes were baking a static composer into the HTML so users could type during React initialization, precompiling a V8 code cache for the desktop shell’s main process, keeping the composer mounted between conversations, prefetching sessions when users hovered over them, and cutting sidebar re‑renders by roughly 90 %. Claude also uncovered layout‑shift jank that existing monitors missed. Using the Layout Instability API, it created a telemetry event that mapped layout‑shift sources to named regions and phases, added an integration test, and used the test as a benchmark to prove fixes. After deployment, field data showed that 31 % of web page loads moved something after the page was usable without user interaction; Claude addressed the top causes in batches and continued iteratively. Over the course of the sprint, more than three thousand changes were merged without any customer‑facing incident or rollback, illustrating a loop where measurement enabled rapid, safe optimization.
Key facts
- Core user experience made about 3x faster
- Time to typeable page fell from 3.1 s to 0.55 s (75th percentile)
- New Claude Code session time fell from 0.8 s to 0.3 s
- Claude Cowork cloud session load time fell from 2.6 s to 0.73 s
- More than 3000 changes merged with zero customer‑facing incidents
- Slack channel tasked Claude with performance monitoring and improvement
- Four key journeys identified, expanded to thirteen measurements
- Static composer baked into HTML, V8 code cache precompiled
- Sidebar re‑renders cut by ~90 %
- Layout‑shift telemetry revealed 31 % of page loads moved after usability
- Tens of thousands of user‑hours saved daily