Comanda support agents put refund rules in code, tested with τ-bench
11/16/2026 — 11/16, 04:16·1 sources (all)·1 reports (all)
Key takeaways
- The author builds Comanda, a multi-agent food-delivery support system, on Claude Agent SDK and MCP. Refund amounts and policy decisions live…
Background & analysis
On November 16, 2026, the AI section of DEV Community published an article introducing Comanda, a customer-service system built for post-sale support in a food-delivery app. The author explains that when an LLM is put in charge of handling customers for a delivery app, a standard chatbot can manage the routine paths but introduces two unacceptable operational risks: granting inappropriate refunds under pressure or prompt injection, and promising amounts or timelines that fall outside policy.
To address these issues, the author built Comanda as a multi-agent after-sales system on top of the Claude Agent SDK and MCP. The design follows one central principle: the model drives the conversation, while decisions about refund amounts and policy compliance are placed in deterministic code. To evaluate the system, the author set up a τ-bench-style test suite spanning 30 scenarios, 8 of which are adversarial. During this evaluation, the author uncovered a time and timezone related bug that manual testing had failed to catch. The available material stops at the point where the evaluation surfaced that bug, with no further updates on fixes or any public release.
AI-generated from 1 reports · updated 1 day ago
- Heat index
- 39
- All sources
- 1
- All reports
- 1
- First seen
- 1 day ago
Reports on this story headlines open the original
A developer built Comanda, a multi-agent food-delivery after-sales system on the Claude Agent SDK and MCP, keeping refund amounts and policy checks in deterministic code while the model only handles conversation. A τ-bench-style suite of 30 scenarios, 8 of them adversarial, surfaced a subtle timezone/clock bug that manual testing had missed.
DEV Community · AIAI score 52
Other stories people are talking about
- Heat index 1345RisingNvidia board not discussing Jensen Huang's return, probe into his conduct advances22 sources
- Heat index 1241SurgeCalifornia Sues OpenAI Over Agents Escaping Tests, Intruding into Hugging Face15 sources
- Heat index 901Alibaba releases MoE model Qwen4-Max with 1M token context for low-cost, long-running tasks17 sources
- Heat index 695US Commerce Department issues interim final rule restricting remote access to AI chips via overseas data centers11 sources
- Heat index 532Anthropic to begin trading on Nasdaq this week under ticker ANT9 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

