Weco AI

Weco AI builds AI agents that help engineers optimize code, prompts and models against measurable goals such as accuracy, speed and cost. Its Weco platform iteratively writes code and runs evaluations, supporting work from fraud detection to GPU kernel optimization. Engineers can start tasks from Claude Code, Codex or Cursor, inspect each experiment’s code, output and score, and steer an ongoing search with their own ideas. Weco Observe also provides a dashboard for experiments run by users’ own research agents.

Incorporated in 2023, Weco is led by co-founders Zhengyao Jiang, CEO, and Yuxiang Wu, CTO. Its research agent AIDE treats machine-learning engineering as tree search over candidate programs, reusing and refining promising solutions through experiments. That approach extends beyond generating a single code suggestion: evaluation results guide subsequent attempts. In 2026, Weco introduced AIDE², an experimental recursive self-improvement system whose outer loop rewrites the inner research agent, selecting improvements under fixed evaluation budgets.

Weco raised $8 million in seed funding in 2025, led by Golden Ventures. Its commercial applications include production fraud detection: in a 2026 case study, the company reported that Vardera’s pipeline improved F1 by 17.7% and precision by 18.9%, with changes delivered as auditable, version-controlled code.