Can Claude Code Build a Year‑Long System in Just One Hour?

Claude Code reproduces a year‑long Google project in one hour

Google senior engineer Jaana Dogan reported that a three‑paragraph prompt given to Anthropic’s Claude Code generated a functional prototype of a distributed agent orchestrator in about one hour—very close to the system her team spent a year building internally. The prompt contained no proprietary details; it described a “toy version” of the intended architecture.

Usage constraints

Within Google, Claude Code is currently permitted only for open‑source projects and cannot be used on internal codebases.

Evolution of LLM‑assisted coding (as observed by Dogan)

2022 : single‑line code completion

2023 : generation of complete code snippets

2024 : ability to work across multiple files and build simple applications

2025 : creation and refactoring of entire codebases

Technical workflow recommended by Claude Code creator Boris Cherny

Key steps for effective use of Claude Code:

Start tasks in Plan mode to define requirements and architecture before code generation.

Use slash commands and sub‑agents to automate repetitive actions such as code simplification, end‑to‑end testing, or documentation generation.

For longer tasks, run background agents that produce intermediate results, then have a separate Claude instance review and refine the output.

During code review, @Claude in pull requests to add missing documentation or enforce style rules.

Integrate Claude with external tools (e.g., Slack, BigQuery, Sentry) to embed AI‑generated code into the broader engineering pipeline.

Productivity insights and caveats

Several engineers, including former Google and Meta staff, note that AI excels at automating repetitive coding work but does not replace the high‑level problem definition, system design, and alignment effort that dominate a project's timeline. The real time savings come from reducing manual coding, not from eliminating meetings, architectural debates, or specification work.

There is concern that organizations may interpret these efficiency gains as justification for workforce reductions rather than redeploying engineers to higher‑value tasks.

Google‑Anthropic partnership context

Google holds roughly 14 % of Anthropic and has invested about $3 billion. In October 2025 the companies agreed that Google could provide up to one million TPUs to Anthropic, a commitment expected to deliver over 1 GW of compute capacity by 2026.

Claude Code contribution statistics (December 2025)

Boris Cherny disclosed that in a 30‑day period he contributed 259 pull requests and 497 commits (≈40 k lines added, ≈38 k lines removed), all generated with Claude Code paired with the Opus 4.5 model, without opening an IDE.