Source: Nextgov/FCW. Summary and analysis by S2V Advisory.
What happened
In a Nextgov/FCW commentary, KJ Lian of Amazon Web Services and Anil Chaudhry of the Department of Transportation described a pilot run in the ATARC Agentic AI Lab. Three specialized AI agents evaluated a fictional $8.5 million federal proposal for data modernization, each focused on a different domain: Federal Acquisition Regulation compliance, executive orders and technical evaluation. The agents identified gaps in small business subcontracting documentation, security framework details and cost justification, while people retained responsibility for all final determinations. The authors called for testing the approach on real procurement workloads and for policy guidance on AI-assisted document review with human-in-the-loop standards.
What it means for buyers
Although the pilot focused on government acquisition, the model applies directly to commercial procurement: divide a complex review into specialized agents, and keep people accountable for the decision.
Proposal and bid evaluation is one of the most time-consuming parts of sourcing. Evaluators cross-check long submissions against requirements, terms and compliance obligations, and inconsistency between evaluators is common. Specialized agents that each check one dimension, such as commercial terms, compliance or technical fit, can make reviews faster and more consistent, and produce a clear record of what was checked.
Three design choices make the difference:
- Clear evaluation criteria written before bids arrive. Agents apply criteria; they shouldn’t invent them.
- Traceable findings. Every flag should point to the specific passage and requirement behind it, so reviewers can verify it quickly.
- Defined human accountability. Award decisions, and the rationale behind them, stay with the evaluation team.
This is the same principle behind our approach to AI in procurement: AI accelerates the method, and people own the decisions. See our Execute service.