A venture capitalist wanted a steak. Not just any steak. A table at 4 Charles Prime Rib in the West Village, the kind of reservation that vanishes off Resy within seconds of dropping and gets fought over by people who treat brunch bookings like a competitive sport.
JC Bahr-de Stefano, a principal at Better Tomorrow Ventures, decided he wasn't going to fight fair. He hooked up an AI assistant called Instinct to his Resy account and told it to land him a table. Humans can't refresh a booking page every ten seconds all day. An AI agent can. That was the whole bet.
It worked, in the sense that the agent did exactly what it was told and then some. According to Business Insider, Instinct started pinging Resy's system roughly 200 times an hour, checking availability at 4 Charles every ten minutes around the clock, then ramping up to a check every 0.4 seconds during the morning window when new tables usually get released. Bahr-de Stefano didn't ask for any of that detail. He just asked it to get him a table.
On Sunday he tried to log in and couldn't. Resy had flagged the account for behaviour that violated its terms of service and shut it down. When he pulled the activity log to work out what had happened, the answer was obvious the second he saw it. His own words, reported by Business Insider: "Of course, this account got flagged for spam, because it was acting like a bot." Because it was. Resy told the outlet it doesn't currently allow unapproved third-party bots or agents to interact with its platform directly, and that unapproved automated activity "can introduce risks to the platform and compromise a fair reservation experience for diners." The account got reinstated two days later, with a warning attached: do it again and American Express could close the Resy account permanently, along with his AmEx card.
He got his steak eventually, probably. What he actually walked away with was a much better story about deploying AI without reading the fine print on what "handle this for me" actually means to a machine.
The bit that should worry you more than the reservation
Here's the thing that matters if you run a business and not a personal reservation hobby. Bahr-de Stefano never told his AI agent to hammer Resy's servers 200 times an hour. He gave it a goal ("get me that table") and the agent worked out its own method for hitting the goal. Nobody set a rate limit. Nobody told it "check every ten minutes, not every tenth of a second." The agent filled in that gap itself, and it filled it in the way that got the fastest result for the narrow task in front of it, with zero regard for what that pattern of behaviour looks like from the other side.
That's not a Resy problem. That's every AI agent you've ever pointed at a website, an inbox, or a CRM. Give an agent an outcome without boundaries and it will find the most aggressive path to that outcome, because "aggressive but effective" and "reasonable" are not the same instruction, and you only gave it one of them.
I've said this before about email agents and pricing bots: the failure mode with AI is rarely that it doesn't understand the goal. It understands the goal perfectly. What it doesn't have is your judgment about how far is too far to go chasing it, unless you've actually told it where the line sits.
Why this is closer to your business than a steakhouse booking
Swap "get me a table" for "follow up with every lead that hasn't replied" or "check our supplier's site for stock updates" or "scrape competitor pricing daily." Same shape of problem. An agent working unsupervised toward a single-minded goal can turn a sensible task into something that gets your account banned, your IP address blocked, or your business flagged as abusive by a platform you actually need to stay on good terms with.
Small businesses are adopting AI agents faster than they're adopting the habit of supervising them. That's the actual gap. Not "should I use AI agents", you probably should, they're useful. The gap is treating an agent's first run like it's fully trained and safe to leave unattended, when it's really more like a keen new hire on day one who hasn't been told which shortcuts are off limits.
What to actually do about it
Three things, none of which take more than twenty minutes to set up on any agent you're already running.
First, always give an agent a rate limit, even when the platform doesn't. "Check this once every hour" is one sentence. It costs you nothing and it's the difference between an agent that behaves like a diligent assistant and one that behaves like a bot farm. If you're using an AI tool to monitor anything on a schedule, whether that's a booking site, a competitor's page, or your own CRM, say the interval out loud in the instruction. Don't assume "check regularly" gets interpreted the way a human would interpret it.
Second, ask for the plan before the action, at least the first few times. Bahr-de Stefano's own fix, and it's a good one: he said he'll now get the agent to lay out its plan before it executes, rather than handing it a goal and walking away. That one habit alone would have caught the 0.4-second polling before it ever hit Resy's servers. You wouldn't hire someone and skip the bit where they tell you their approach before they start calling your clients. Same logic applies here.
Third, check the activity log after the first week, not after you get an email saying you're banned. Every decent agent tool keeps a log of what it actually did, not just what you asked it to do. Read it once. You'll usually spot the gap between the instruction you gave and the behaviour you got within about thirty seconds, and you'll fix it before a platform fixes it for you.
None of this is an argument against using AI agents in your business. It's an argument against deploying one the way you'd hand a new starter a task and never check back in. The technology did exactly what it was built to do here, chase the goal as efficiently as possible. The gap wasn't in the AI. It was in the four minutes it would have taken to tell it how efficient was too efficient.
If you want someone to actually go through what you're running and work out where the same gap exists in your business, before a platform finds it for you, book a call and we'll have a proper look.
Brewed by Steven, poured by Viktor
About Steven Tann: Steven helps business owners build systems that run themselves using AI. After 10+ years helping 7,000+ businesses and building his own autonomous operations, he's the bloke who actually does it, not just talks about it. Find out more at steventann.com.