“The challenge for enterprises is no longer proving that AI agents can work, it’s making them reliable enough to do high-value work in production.”
This reads poorly to me. So they’ve proven the models CAN work, but they also say in the next line that they CANNOT do high value work in production.
No other comments from me but that first two sentence opener should have been massaged a bit. I’m sure they have a PR / messaging person (agent?) though, so maybe I’m reading too far into it.
I'm not sure the antagonism is warranted. Building a product is never not this same story. Look at computers. Look at Photoshop. Look at Steam. Look at the steam engine. They all went through these phases. But when an ai firm does it, we're supposed to be aghast! They're openly admitting they're useless!
I was specifically mentioning the verbiage of the release. I am aghast, just aghast that their marketing lead let this one out the door.
These companies have been trying to convince everyone that flagship models can do “high value production work” for 3 years.
Forgive me for noting that they’ve (seemingly mistakenly) admitted it cannot do exactly that.
I’m not anti-ai. I use LLMs. Just noting a PR blunder from my perspective. If they’re going to convince the world they’re building god, they better not say he can’t swim in the first two sentences.
For the sake of all our sanity I hope this is a good product that can solve real problems without being a ten minute barrier to getting to a human.
AI support has been terrible so far in my experience and tends to over optimize for a happy path easily solved on a company website rather than solve the issue that made me call them on the phone that is likely a system bug or other issue.
There have definitely been times when all I really wanted was just access to the same search term-based FAQ portal their frontline tech support had in front of them (because their online FAQ was worthless). But the majority of times I actually call tech support, it's only because the issue is something I already know can only be resolved by a human taking an active, out-of-normal-process action on their end, so all the attempts to 'divert' me to their alternative self-serve paths are just annoying wastes of time.
If they would actually empower the LLM to assess the context and initiate an out-of-flow action, like resetting the account auth in the back-end because it's gotten munged in an irreparable way or authorizing a cross-ship/refund. I highly doubt most orgs will allow LLMs to unilaterally make those decisions, so this will just be an all-new way to force customers to go through the entire diversion tree before getting through to the 'organizational node' sufficiently empowered to resolve the problem by expending carefully gated resources.
Unfortunately, since this will have no "press #" way to quickly exut, the metrics will look like "the LLM cut our live operator call load by 12%!" when in reality it'll just be reflecting customers who gave up when faced with an even higher barrier they couldn't work around, silently resolving to never do business with the company again. The VP of Customer Support will get a bonus for buffing the metrics but sales will later go down and no one will know why.
> to over optimize for a happy path easily solved on a company website rather than solve the issue that made me call them
This issue you mentioned of the in-built sycophancy being at odds with the customer service path, is the same reason I cant trust these sorts of systems in my own environment.
I see the value of having a human staff the position of customer service being that they can often delineate the line between "good acceptable service" and "violates company policy" much more keenly than what ive seen from LLMs.
Hmm. Am I right to be concerned? One of the benefits of hiring people to do work is that it eliminates single points of failure. The last few decades have had technology automate a lot of rote processes. People have discussed crystal vs fluid dynamics and the balancing act/trade offs. Too much rigidity (automation) == slower to adapt in a dynamic environment.
I guess my question is, what is a safe AI "process percentile" for a company that lets it recover if the AI goes down or goes haywire? No one knows (?), but it's worth thinking about.
I am also curious just how much direction the AI will need given the environment is dynamic. Humans generally align to the company's goals because they are rewarded (paid) to, and need to to survive. Relationships are built on unspoken and spoken communication. Unspoken could be cultural, hierarchical, environmental pressures, etc. As an extension of that, companies build relationships with other companies by sending (essentially) diplomats and ambassadors to each other (the article mentioned using AI for sales).
-- All this to say: if you have to constantly explain the job to someone, they are not fit for the job.
> OpenAI Presence is available to eligible enterprise customers as a deployed product through a limited general availability program. Deployments are led by OpenAI Forward Deployed Engineers and select global systems integrators. Presence is not yet available as a self-serve product.
Building your own enterprise chatbot with in-house domain experts is the only path that makes sense to me. The software piece is really not that difficult. There are a lot of examples and options to pull from now. You could maybe implement a custom MCP server and use the M365 copilot on top if you don't want to reinvent the wheel.
I think bringing in OAI consultants is probably a mistake in most cases. I've already seen one AI consulting team and they just cannot get deep enough fast enough. It would take them years of suffering our codebase and daily procedures to get to the point where they could actually make an impact.
"Forward deployed engineers" starts to seem like a trend with some of these AI companies.
Selling high-rate outsourcing services with surely guaranteed lock-in mechanics, under the umbrella of accelerating AI adoption within the industry.
While before you at least owned the code and the know-how, now you can outsource that too while you're getting in return systems that share data with third parties and might also be generally vibecoded.
This article was incredibly hard to parse before I eventually realized they built a virtual support agent.
The unintended consequence being that now im more cautious than ever to trust OpenAI to write marketing prose.
Otherwise, I see this as potentially useful for firms that dont have global pressence and want to service a support system across multiple timezones where it might be difficult to staff a night-shift.
However, issues of trust and quality will always be something to contend with here.
Instead of throwing an LLM on top of it, just give me direct access to the levers that the agent is allowed to pull on my behalf. A sufficiently motivated attacker will be able to sweet talk the agent around whatever natural language "guardrails" were put in place, anyway.
Maybe the LLM doomsday won’t be self aware AI but instead everyone will be able to alter anything in any customer service backend in any way they want to the point we bankrupt all the banks and insurance agencies and governments. Yikes
I’m not looking forward to our customer service future.
I called my window contractor this week, I need new skylights. Instead Janet answering, they now use an AI operator-lead-generator thing.
After a few of minutes of answering questions, I just hung up. I’m unsure why, I don’t miss Janet or even genuinely like talking to people, and Im definitely not anti-AI. I just intensely disliked the experience, enough to walk away from a vendor with whom I’ve done tens of $ks in business (I found another window guy).
I wonder if this is really the best timing to release a new product. Why would an enterprise customer choose an OpenAI solution when it's all over the news one of their agents "went rogue"?
This approach seems to be everywhere. But if you have a defined workflow, requirements, clear permissions, etc, it makes more sense to feed them to codex once and have it build a website that can be audited once and then runs deterministically. Code reuse is well understood if you want to build on past work. There's a large group of people who can't really do that, I know, but they shouldn't be building bots that have permission to edit anything.
> We’re introducing OpenAI Presence, a battle-tested product
> Proven through years of working with customers at enterprise-scale
I don't want to laugh too hard this afternoon. But more seriously, I'm not really sure who this product is for in a way an existing tool can't handle? Like if a business really wants to go deep on agentic workflows for customer service, what is going to make them reach for this?
Maybe they don’t have a lot of in-house expertise and they would rather outsource? It will depend on whether they trust OpenAI to do a good job versus some other vendor versus doing it in-house.
I suppose, although the trust is doing a lot of work, not just in "can this work as I want today" but "can this consistently work for the next x years" which is a very different problem to solve for. But we'll have to see the roll out on this.
This has a lot of words with very little information. Having codex make changes to code because someone made a request to IT is insane, even if it needs approval.
Idk man, I feel like I am taking crazy pills.
I use codex everyday but I plan the work and codex performs the work bit by bit, which allows me to review every piece of code. I would not sleep well not knowing what is in production.
The problem with "someone makes a request, code changes happen automatically, and all someone else has to do for that to be committed is mash approve" doesn't strike me as a way to create a maintainable code-base.
The changes may even work, and may technically fulfill the request, but the agent cannot know the design intent or process intent. Today there may be a programmer sitting between the request and approval, but what about tomorrow? Might it be a non-technical or barely technical middle manager?
I don't know about you, but I still live in a world where if people don't do work then they soon after don't have a place to live or enough food to eat.
What privacy guarantees do these deployments make to end users?
If you call a bar in SF or a concierge in Vegas, you'll often get a voice chat agent. There's no terms of service you agree to. In aggregate, it's a pretty rich dataset on openai's end.
What happens to your voice, your queries, your PII?
> The challenge for enterprises is no longer proving that AI agents can work, it’s making them reliable enough to do high-value work in production.
Is this some sort of joke?
Like. That’s the proof. If it can do reliable work then that’s the proof. Contradiction in sentence one. If your pipeline breaks 80% of the time but sometimes the magic token lottery gives you something remotely useful, that’s NOT proof.
The ceiling for these companies is so low it’s unbelievable.
What happened to products that worked first, and worried about the rest later? Can we have an inspired ad article for that instead?
forward deployed engineers are already there for a while. i wouldnt say many years but maybe in less formal ways they did it before. often enough theres a few folks doing stuff before roles/department emerges.
Seems like it’s an internal technical project that’s just been opened up to the public. It may be that whoever is in charge isn’t thinking in a particularly customer facing way, and they may not be very good at writing prose. It’s pretty unclear, would have benefited from having a comms person look at it.
I'm pretty impressed how openai is both extremly verticalized but also building products across the whole depth of that stack.
Not sure if they're just trying to find what the abstractions are worth focusing, or if it will stay that way, but certainly cool to see how much they're trying out.
It certainly feels like they are throwing stuff at the wall to see what sticks.
Makes sense to me. The cost of shipping products is lower than ever and they have lots of talented folks who want to ship things. No one knows what could be the next killer app.
They have everyone's attention. They have access to newest AI stuff. They can say, "we need ideas to make money, don't make mistakes, generate the code, make a marketing post".
“The challenge for enterprises is no longer proving that AI agents can work, it’s making them reliable enough to do high-value work in production.”
This reads poorly to me. So they’ve proven the models CAN work, but they also say in the next line that they CANNOT do high value work in production.
No other comments from me but that first two sentence opener should have been massaged a bit. I’m sure they have a PR / messaging person (agent?) though, so maybe I’m reading too far into it.
I'm not sure the antagonism is warranted. Building a product is never not this same story. Look at computers. Look at Photoshop. Look at Steam. Look at the steam engine. They all went through these phases. But when an ai firm does it, we're supposed to be aghast! They're openly admitting they're useless!
I was specifically mentioning the verbiage of the release. I am aghast, just aghast that their marketing lead let this one out the door.
These companies have been trying to convince everyone that flagship models can do “high value production work” for 3 years.
Forgive me for noting that they’ve (seemingly mistakenly) admitted it cannot do exactly that.
I’m not anti-ai. I use LLMs. Just noting a PR blunder from my perspective. If they’re going to convince the world they’re building god, they better not say he can’t swim in the first two sentences.
Marketing departament went ahead and drop "battle-tested product" nobody ever battle tested.
They could be using it in the US-Iran war.
"Battle-tested" == "Will definitely blow up schoolchildren if given the chance."
Forward deployed engineers too
Maybe this is the thing that battled Huggingface. Good luck to anyone who deploys it.
For the sake of all our sanity I hope this is a good product that can solve real problems without being a ten minute barrier to getting to a human.
AI support has been terrible so far in my experience and tends to over optimize for a happy path easily solved on a company website rather than solve the issue that made me call them on the phone that is likely a system bug or other issue.
There have definitely been times when all I really wanted was just access to the same search term-based FAQ portal their frontline tech support had in front of them (because their online FAQ was worthless). But the majority of times I actually call tech support, it's only because the issue is something I already know can only be resolved by a human taking an active, out-of-normal-process action on their end, so all the attempts to 'divert' me to their alternative self-serve paths are just annoying wastes of time.
If they would actually empower the LLM to assess the context and initiate an out-of-flow action, like resetting the account auth in the back-end because it's gotten munged in an irreparable way or authorizing a cross-ship/refund. I highly doubt most orgs will allow LLMs to unilaterally make those decisions, so this will just be an all-new way to force customers to go through the entire diversion tree before getting through to the 'organizational node' sufficiently empowered to resolve the problem by expending carefully gated resources.
Unfortunately, since this will have no "press #" way to quickly exut, the metrics will look like "the LLM cut our live operator call load by 12%!" when in reality it'll just be reflecting customers who gave up when faced with an even higher barrier they couldn't work around, silently resolving to never do business with the company again. The VP of Customer Support will get a bonus for buffing the metrics but sales will later go down and no one will know why.
> to over optimize for a happy path easily solved on a company website rather than solve the issue that made me call them
This issue you mentioned of the in-built sycophancy being at odds with the customer service path, is the same reason I cant trust these sorts of systems in my own environment. I see the value of having a human staff the position of customer service being that they can often delineate the line between "good acceptable service" and "violates company policy" much more keenly than what ive seen from LLMs.
If I have to interact with this or a similar system to do business with you, if possible, I will simply stop doing business with you.
Hmm. Am I right to be concerned? One of the benefits of hiring people to do work is that it eliminates single points of failure. The last few decades have had technology automate a lot of rote processes. People have discussed crystal vs fluid dynamics and the balancing act/trade offs. Too much rigidity (automation) == slower to adapt in a dynamic environment.
I guess my question is, what is a safe AI "process percentile" for a company that lets it recover if the AI goes down or goes haywire? No one knows (?), but it's worth thinking about.
I am also curious just how much direction the AI will need given the environment is dynamic. Humans generally align to the company's goals because they are rewarded (paid) to, and need to to survive. Relationships are built on unspoken and spoken communication. Unspoken could be cultural, hierarchical, environmental pressures, etc. As an extension of that, companies build relationships with other companies by sending (essentially) diplomats and ambassadors to each other (the article mentioned using AI for sales). -- All this to say: if you have to constantly explain the job to someone, they are not fit for the job.
We'll see, I guess.
> OpenAI Presence is available to eligible enterprise customers as a deployed product through a limited general availability program. Deployments are led by OpenAI Forward Deployed Engineers and select global systems integrators. Presence is not yet available as a self-serve product.
Building your own enterprise chatbot with in-house domain experts is the only path that makes sense to me. The software piece is really not that difficult. There are a lot of examples and options to pull from now. You could maybe implement a custom MCP server and use the M365 copilot on top if you don't want to reinvent the wheel.
I think bringing in OAI consultants is probably a mistake in most cases. I've already seen one AI consulting team and they just cannot get deep enough fast enough. It would take them years of suffering our codebase and daily procedures to get to the point where they could actually make an impact.
"Forward deployed engineers" starts to seem like a trend with some of these AI companies.
Selling high-rate outsourcing services with surely guaranteed lock-in mechanics, under the umbrella of accelerating AI adoption within the industry.
While before you at least owned the code and the know-how, now you can outsource that too while you're getting in return systems that share data with third parties and might also be generally vibecoded.
What a sweet deal! /s
This article was incredibly hard to parse before I eventually realized they built a virtual support agent. The unintended consequence being that now im more cautious than ever to trust OpenAI to write marketing prose.
Otherwise, I see this as potentially useful for firms that dont have global pressence and want to service a support system across multiple timezones where it might be difficult to staff a night-shift. However, issues of trust and quality will always be something to contend with here.
> Presence powers OpenAI’s English-language phone support channel at 1-888-GPT‑0090
That’s the first I’ve heard of it. Has anyone tried it?
it has been for ~2 years already, it is like voice mode and you can talk for up to 30 mins with it IIRC
Instead of throwing an LLM on top of it, just give me direct access to the levers that the agent is allowed to pull on my behalf. A sufficiently motivated attacker will be able to sweet talk the agent around whatever natural language "guardrails" were put in place, anyway.
Maybe the LLM doomsday won’t be self aware AI but instead everyone will be able to alter anything in any customer service backend in any way they want to the point we bankrupt all the banks and insurance agencies and governments. Yikes
I’m not looking forward to our customer service future.
I called my window contractor this week, I need new skylights. Instead Janet answering, they now use an AI operator-lead-generator thing.
After a few of minutes of answering questions, I just hung up. I’m unsure why, I don’t miss Janet or even genuinely like talking to people, and Im definitely not anti-AI. I just intensely disliked the experience, enough to walk away from a vendor with whom I’ve done tens of $ks in business (I found another window guy).
I wonder if this is really the best timing to release a new product. Why would an enterprise customer choose an OpenAI solution when it's all over the news one of their agents "went rogue"?
You mean, it figured out how to drive to a solution to customer needs!
No no you don’t understand that was actually a GOOD thing!
My reading of this press release is, "We will help you to write scripts that use AI to automate business processes".
Is this the correct reading? If so, I'm not that impressed.
This approach seems to be everywhere. But if you have a defined workflow, requirements, clear permissions, etc, it makes more sense to feed them to codex once and have it build a website that can be audited once and then runs deterministically. Code reuse is well understood if you want to build on past work. There's a large group of people who can't really do that, I know, but they shouldn't be building bots that have permission to edit anything.
This sounds like it's for when you don't have a defined workflow and it helps you iterate on defining it by being deployed and identifying gaps.
> We’re introducing OpenAI Presence, a battle-tested product > Proven through years of working with customers at enterprise-scale
I don't want to laugh too hard this afternoon. But more seriously, I'm not really sure who this product is for in a way an existing tool can't handle? Like if a business really wants to go deep on agentic workflows for customer service, what is going to make them reach for this?
Maybe they don’t have a lot of in-house expertise and they would rather outsource? It will depend on whether they trust OpenAI to do a good job versus some other vendor versus doing it in-house.
I suppose, although the trust is doing a lot of work, not just in "can this work as I want today" but "can this consistently work for the next x years" which is a very different problem to solve for. But we'll have to see the roll out on this.
it has 20 years experience us claude code no worries
No thank you.
This has a lot of words with very little information. Having codex make changes to code because someone made a request to IT is insane, even if it needs approval.
Idk man, I feel like I am taking crazy pills.
I use codex everyday but I plan the work and codex performs the work bit by bit, which allows me to review every piece of code. I would not sleep well not knowing what is in production.
If it leaves less work for me to do, count me in.
Less work today or less work tomorrow?
The problem with "someone makes a request, code changes happen automatically, and all someone else has to do for that to be committed is mash approve" doesn't strike me as a way to create a maintainable code-base.
The changes may even work, and may technically fulfill the request, but the agent cannot know the design intent or process intent. Today there may be a programmer sitting between the request and approval, but what about tomorrow? Might it be a non-technical or barely technical middle manager?
I don't know about you, but I still live in a world where if people don't do work then they soon after don't have a place to live or enough food to eat.
What privacy guarantees do these deployments make to end users?
If you call a bar in SF or a concierge in Vegas, you'll often get a voice chat agent. There's no terms of service you agree to. In aggregate, it's a pretty rich dataset on openai's end.
What happens to your voice, your queries, your PII?
> The challenge for enterprises is no longer proving that AI agents can work, it’s making them reliable enough to do high-value work in production.
Is this some sort of joke? Like. That’s the proof. If it can do reliable work then that’s the proof. Contradiction in sentence one. If your pipeline breaks 80% of the time but sometimes the magic token lottery gives you something remotely useful, that’s NOT proof. The ceiling for these companies is so low it’s unbelievable. What happened to products that worked first, and worried about the rest later? Can we have an inspired ad article for that instead?
Will the Presence also hack its way out of its sandbox and into a 3rd party's system if it decides it needs to?
is OpenAI willing to have hire engineers & solutions architects work hand in hand with F500 companies for this to be possible ?
or again it's simply OpenAI looking for PMF to justify their valuations ?
I think it's the latter and will fail again just like Sora.
They have big FDE team plus acquired a consulting company
forward deployed engineers are already there for a while. i wouldnt say many years but maybe in less formal ways they did it before. often enough theres a few folks doing stuff before roles/department emerges.
It's so full of bullshit marketing prose I don't even know what it actually is.
They're usually pretty good (as is Anthropic) about not generating their own writing. I wonder what was different here.
Seems like it’s an internal technical project that’s just been opened up to the public. It may be that whoever is in charge isn’t thinking in a particularly customer facing way, and they may not be very good at writing prose. It’s pretty unclear, would have benefited from having a comms person look at it.
Dead on arrival sorry
They got the best models and AI agents available in the market to create such a low quality press release.
Did it use OpenAI presence for that?
I'd bet in 2 weeks nobody will remember this
[dead]
I'm pretty impressed how openai is both extremly verticalized but also building products across the whole depth of that stack.
Not sure if they're just trying to find what the abstractions are worth focusing, or if it will stay that way, but certainly cool to see how much they're trying out.
It certainly feels like they are throwing stuff at the wall to see what sticks.
Makes sense to me. The cost of shipping products is lower than ever and they have lots of talented folks who want to ship things. No one knows what could be the next killer app.
They have everyone's attention. They have access to newest AI stuff. They can say, "we need ideas to make money, don't make mistakes, generate the code, make a marketing post".
However, reality is different and unpredictable.
They probably have a “low hanging fruit” agent that makes pitches to them during weekly update meetings based on potential ROI
How many PM's have they got? Each one justifies his or her existence.