>"The pattern was every morning and evening during rush hour, there would be an accident, and hundreds of callers called in about the same thing."
>"Anyone calling 911 within a 200-meter radius of a live auto accident is now immediately routed to the AI agent"
>"the tool engages only for motor vehicle accident calls, only when every human call taker is already busy, & only when the caller is within about 200 meters of a crash that has already been reported."
How does "the tool" know its a motor vehicle accident call, and how does it know when the caller is within 200 meters of an already reported crash? Seems like the screening will have to happen on all calls.
The key part being it’s only when there is a surge and queue of calls. Crash calls will most likely 99% of the time if not 100 be from cell phones. I am guess the AI agent will ask for a description of the call.
I have seen some AI call tech before and it can work really well. Obviously depends on their implementation but imo a great use of technology.
Reasonable would be to play a message "We know about the accident on Elm Street, we're already dispatched a team, if you're calling about that, you may hang up".
No need to waste people's time with an "AI agent".
Service providers have been doing that since forever.
Everything can be handled by playing a message and reading a response with far superior, refined, cheaper and privacy-respecting technology than voice recognition - pushing a button.
presumably so that if you start shouting about the heart attack you are having, that you dont need to wait until the end of the prerecorded message to get routed to a human operator
im not sure if you've ever been in a panic-filled emergency situation and had to dial 911, but the easier and more adaptable you can make it, the better. requiring button pushes, or anything else other than voice, should be avoided if possible.
this comes across as one of the best possible use cases of ai to me: only used in strict circumstances when human operators are busy, with the goal of deduping calls and reducing 911 queues.
and its been working really well, from the looks of it, for 3 years now
I think the impression, that "AI" would make it easier is not only far-fetched, but utterly misguided.
So far it hasn't made anything easier (support call lines are the proof of that).
I understand the picture of that outcome, which you'd like to have, but our real world experience just doesn't support it.
> requiring button pushes, or anything else other than voice, should be avoided if possible.
Well, even movie tropes often show, how that is not the case. You might be able to push a button safely, if the intruder in your house, but will be unable to comply with "I couldn't hear you, could you speak in a loud and clear voice?" without compromising your safety.
> and its been working really well
No caller satisfaction surveys, no comparison with tried and true button approach, just statements from parties with vested interest.
Very reasonable and also a good use of the technology imo. Like all things it depends on the implementation but I can see this adding a lot of value for 911 call centers.
The assumption that because something is important and the consequences are dire when it fails, infinite resources should be allocated is simply wrong. At some point 911 calls will be backlogged. What to do then? Maybe AI triage is the best of the worst options.
Was New Orleans "investing infinite resources" before? It seems like as always the pitch here is you fire a bunch of existing 911 operators and split the projected cost savings with an AI company.
It's also highly likely that the cost savings projections are optimistic and they will be back at the table a year from now saying that the only way to realize the projected savings is to divert more calls to the AI and further reduce staffing.
The assumption is more that, given tools to automate the “less important” parts of important processes, this starts a slippery slope where capitalism will squeeze as many humans as possible out of the job to reduce costs because the “AI is good enough.”
Until, of course, it isn’t - but at that point, the budget is reallocated, there’s no funding, there’s simply nothing we can do except continue to pay this AI contract.
This is a slippery slope, where the number of operators can be allowed to gradually decline until eventually it’s just AI. And given the unreliability of LLMs, this is a grave concern that will undoubtedly lead to the enshittification of critical services.
If you don't give the cops their toys, they might go on a work strike, threaten politicians, or whatever. If you don't give 911 operators money and more seats the backlash is much more muted.
911 calls answered by AI in New Orleans [video] – 2026-08-11 – https://news.ycombinator.com/item?id=49258252
New Orleans is testing Carbyne’s AI-powered Emergency Call Triage software – 2026-08-07 – https://news.ycombinator.com/item?id=49204546
>"Anyone calling 911 within a 200-meter radius of a live auto accident is now immediately routed to the AI agent"
>"the tool engages only for motor vehicle accident calls, only when every human call taker is already busy, & only when the caller is within about 200 meters of a crash that has already been reported."
this seems very reasonable to me
I have seen some AI call tech before and it can work really well. Obviously depends on their implementation but imo a great use of technology.
No need to waste people's time with an "AI agent".
Service providers have been doing that since forever.
no additional time is being wasted compared to a recorded message.
Everything can be handled by playing a message and reading a response with far superior, refined, cheaper and privacy-respecting technology than voice recognition - pushing a button.
presumably so that if you start shouting about the heart attack you are having, that you dont need to wait until the end of the prerecorded message to get routed to a human operator
im not sure if you've ever been in a panic-filled emergency situation and had to dial 911, but the easier and more adaptable you can make it, the better. requiring button pushes, or anything else other than voice, should be avoided if possible.
this comes across as one of the best possible use cases of ai to me: only used in strict circumstances when human operators are busy, with the goal of deduping calls and reducing 911 queues.
and its been working really well, from the looks of it, for 3 years now
So far it hasn't made anything easier (support call lines are the proof of that).
I understand the picture of that outcome, which you'd like to have, but our real world experience just doesn't support it.
> requiring button pushes, or anything else other than voice, should be avoided if possible.
Well, even movie tropes often show, how that is not the case. You might be able to push a button safely, if the intruder in your house, but will be unable to comply with "I couldn't hear you, could you speak in a loud and clear voice?" without compromising your safety.
> and its been working really well
No caller satisfaction surveys, no comparison with tried and true button approach, just statements from parties with vested interest.
It's also highly likely that the cost savings projections are optimistic and they will be back at the table a year from now saying that the only way to realize the projected savings is to divert more calls to the AI and further reduce staffing.
Until, of course, it isn’t - but at that point, the budget is reallocated, there’s no funding, there’s simply nothing we can do except continue to pay this AI contract.
Classic case of “of course they knew what could go wrong, and that is exactly why they chose that solution.”
maybe they need a strong union and strikes?