How Baidu’s AI‑Powered Code Assistant Boosts R&D Efficiency by Over 11 %

In the rapidly evolving business landscape, effective marketing and high‑quality platform services are critical, prompting the need for faster product iteration and AI‑native development to meet user demands.

Baidu's Marketing Service team rebuilt its platform around the Wenxin large model and the intelligent coding assistant Baidu Comate, aiming to accelerate R&D, lower learning barriers, and enhance both user experience and advertising efficiency.

Baidu Comate integrates large‑language‑model technology to provide real‑time code recommendations across 100+ programming languages and 10+ IDEs, covering pre‑, post‑, and mobile‑side development. It delivers code suggestions within 300 ms, enabling engineers to continue typing while the assistant generates continuations.

Adoption within the team is high: over 95 % of engineers use Comate, with generated code accounting for 29.42 % of commits. When a recommendation is accepted, single‑line coding time drops from 16.54 s to 1.74 s, yielding a 35 % efficiency gain (adoption rate 40 %). Even when not adopted, a 7 % efficiency loss occurs, resulting in an overall daily productivity increase of 11.2 % per engineer.

For example, engineers can generate shell scripts from natural‑language prompts in seconds, eliminating the need to search for historical scripts and reducing script‑writing time dramatically.

The Comate Open Platform enables custom extensions, knowledge‑base integration, and RAG‑enhanced code generation. It streamlines multi‑tool workflows; a typical offline environment update that previously required eight platform switches can now be completed with a single command, saving over 10 minutes per deployment and an estimated 467 man‑hours per month.

These efficiencies translate to tangible marketing outcomes: the "Qing Gu" platform processes over 20 k daily conversations, has generated 4 million user interactions, and helped a leading IT education group improve ad conversion by 23.3 % and ROI by 22.7 %.

Looking forward, Baidu plans to deepen AI‑native R&D, expand Comate’s capabilities, and continue exploring large‑model‑driven innovations across the software development lifecycle.