3 Hours with Claude Opus 4.7: functional study webapp and remote MCP- Oneshotted
anthropic claude cohere
| Source: HN | Original article
Claude Opus 4.7 proved its long‑horizon autonomy in a three‑hour live test that produced a fully functional study‑webapp and a remote model‑control panel (MCP) without human‑written code. The developer, working from a single prompt, asked Claude to design the UI, generate a Flask backend, wire up a PostgreSQL database, and expose an API that could be invoked from a separate browser‑based control panel. Within minutes the model delivered a complete project skeleton, and after a brief cycle of clarification prompts it refined authentication, added pagination and deployed the stack to a free Heroku instance. By the end of the session the webapp was live, data could be entered, and the remote MCP allowed the user to toggle model parameters and view token usage in real time.
Why it matters is twofold. First, the test confirms the claims made in Anthropic’s own rollout notes that Opus 4.7 can sustain “hard problems” for hours, a leap from earlier models that frequently stalled after a few hundred tokens. Second, the ability to generate end‑to‑end production code cuts the iteration loop that has limited AI‑assisted development to snippets and prototypes. For startups and enterprises that already face talent shortages, a model that can deliver deployable services on its own could reshape engineering budgets and speed time‑to‑market.
What to watch next includes Anthropic’s upcoming integration of Opus 4.7 into Vertex AI and AWS Bedrock, which will make the model accessible at scale and potentially lower the $5‑$25 per‑million‑token barrier. The community is also testing best‑practice templates that pair detailed plans with “high‑effort” prompts, a technique highlighted in our earlier analysis of Opus 4.7’s performance on April 18. Follow‑up benchmarks against Sonnet 4.8 and Mythos 5 will reveal whether Opus’s autonomy translates into consistent quality across domains, and whether developers will adopt it as a primary coding partner or keep it as a niche assistant.
Sources
Back to AIPULSEN