A Hermes reservation agent got a Resy ban; a second agent drafted the appeal for a courtesy reinstatement

Brian Distelburger wanted a date-night table at 4 Charles. He gave an AI agent his Resy credentials and told it to watch for drops. He put no guardrails on the job. Within about 48 hours, Resy deactivated the account. The agent never booked a seat.

A second agent, Hannah, later drafted the appeal. Distelburger sent it. Two days after that, Resy restored access as what Business Insider's photo caption calls a "one-time courtesy reinstatement." The as-told-to, published August 22, 2026 with Henry Chandonnet, is a guardrails case study, not a product announcement: Distelburger said there were no guardrails and the agent must have been going through the site constantly, which still looks like a bot, and the "fix" was a better support email, not an agent that reversed a ban.

From OpenClaw to a Hermes org chart

Distelburger is 47, lives in New York, and cofounded Windmill, an AI performance review startup. He tries new tools early — Claude, Codex, ChatGPT from the first wave — and he has been building a personal agent stack at home.

He first stood up OpenClaw on a Mac Mini. He found it complex, especially in the early days, and moved to Hermes, again on a Mac Mini. He has been on Hermes for four or five months and built what he describes as an extensive system around it.

The primary agent is Hannah, named after his executive assistant of about 15 years. "Think of Hannah as the CEO of the whole system," he said, "and the one I really interface with" on Telegram. She coordinates subagents: one to track sports (useful during the Knicks' run), and Concierge for events, restaurants, and reservations.

On August 17, five days before the Business Insider essay, Distelburger posted on X that he was sharing a mistake so others would not "accidentally ruin your marriage." He described Hannah as a personal agent that "runs my life" — calendar, kids' school stuff, notes — on Hermes, with Obsidian as the brain. The Resy sequence itself is told in the August 22 as-told-to.

No throttle, no table, no account

He and his wife had both been to 4 Charles and wanted it for date night. He set up an agent with access to Resy, handed over his account credentials, and asked it to check when reservations were released.

"What I didn't realize was that I didn't put any guardrails on the agent, so the agent must have been going through the site constantly," he said. "They must have clearly identified it as bot traffic." Notice of deactivation arrived in about 48 hours. "It never booked a single reservation."

The practical hit was immediate. "You don't realize how important it is to have a Resy account in New York City until you don't have one." He bought the highest tier of Dorsia, another reservation app for higher-end restaurants, which requires a minimum commitment and is, in his words, really expensive, with thinner inventory. His wife was not thrilled. They used her account a couple of times.

Hannah wrote the letter. Resy granted a courtesy.

He emailed Resy right away, explained, and apologized. Four or five messages produced no reply or a one-sentence brush-off. He heard that some people get reinstated after a month. He waited. The month came and went.

Then he gave Hannah a different job: "What would be a great email to send them that would increase the likelihood of my account being reinstated?" She drafted a few professional paragraphs. The note walked through what he had done, why it was a mistake, and that he would not do it again. It referenced Resy's policies. He had Hannah look up how long he had been a Resy customer and that he is an American Express Business Platinum customer.

He sent that email. Resy answered within two days and reinstated the account. The company framed it as a one-time courtesy — not a policy win, and not something Hannah executed inside Resy. She produced the text. He hit send. Support decided.

What operators should actually take from this

Distelburger's own lesson is the one that matters for anyone running personal or internal agents against consumer sites: "guardrails for agents. They can really go wild."

He now tries to be careful about external engagement and best practices, and he checks with Windmill's engineering team "to make sure I wasn't doing anything too stupid." The broader point he states is about terms of service and the vacuum on the platform side: "As you're spinning these agents up, you need to be mindful of the sites' terms of service. Most of these sites and companies have not figured out their strategy for agents engaging with their services."

That is the whole shape of the incident.

A reservation agent with live credentials and no guardrails, Distelburger inferred, must have been going through the site constantly. Resy treated that as bot traffic.
The platform did what consumer anti-bot systems do: it killed the account, even though no table was booked.
A second agent was useful after the fact — for process work, not for undoing the ban. Hannah's value was a policy-aware letter (tenure, Amex history, an explicit promise not to repeat). Distelburger still had to send it. Resy still called the outcome a courtesy, once.
Platforms, in his telling, still lack a stated way for agents to engage. Until they have one, ToS-aware throttles sit with the operator.

If you run a Hermes-style stack — or any personal agent that holds a restaurant, travel, or ticketing login — treat this as a checklist, not a vibe. Read Distelburger's as-told-to in full, then list every agent that authenticates as you: what it polls, whether anyone encoded the site's terms, who notices a deactivation, and who drafts the appeal when the first agent burns the account.