- Decision APIs are small, fast AI models that return structured answers
- Nearby Computing CTO David Carrera says decision APIs could finally give telcos the edge computing use case they've been looking for
- AvidThink's Roy Chua is skeptical about the edge but says there's plenty of reason for telcos to tinker with these models
Decision APIs went from obscure to everywhere in a matter of weeks. But in a world flooded with AI noise, why should telcos care?
In a nutshell, a decision API is a small, fast AI model built to return a quick, structured answer instead of a long, free-form response.
Nearby Computing CTO David Carrera told Fierce these mini-models could finally provide the edge use case telcos have been looking for. AvidThink Founder Roy Chua said he has his doubts about the edge but urged telcos to pay attention to the buzz regardless.
The use cases are real, the demos are impressive, but where the models will run – at the edge, on-device/on-premise or in regional data centers – remains to be seen.
“I’m still slightly skeptical, but it is something they should look at. They should not ignore it,” Chua told Fierce.
What is a decision API, anyway?
Decision APIs are still very new phenomenon, having made their debut in September when TypeSafe AI released its Jev model. Within six weeks, Microsoft, AWS, OpenAI, Cloudflare and Perplexity all released their own decision API models.
They’re different from large language models in three major ways: they’re smaller (anywhere from a few hundred million to a few billion parameters), they’re faster and they make structured decisions (think classification, routing, grading or any other decision where the answers are finite).
“Decision models give you this advantage of giving you a structured output that’s really fast,” Chua said.
Simple use cases include sorting email, triaging incoming support requests, routing between models, and gating AI agents by checking whether an action is safe and permitted. But their speed is why people are also eyeing their use for physical AI, smart buildings, industrial controls and network configuration, such as deciding how much traffic should go where.
What does it mean for the edge?
To hear Nearby Computing co-founder and CTO David Carrera tell it, decision APIs could be the edge breakthrough telcos have been waiting for.
“I think the use case that we were looking for for edge computing in the past may be coming from these decision APIs, where you will have these models everywhere,” Carrera told Fierce. “We truly believe that this is something that will be a game changer.”
He argued that agent-driven applications depend heavily on large language models, and that time to token determines whether a system feels interactive or slow. Decision APIs, he said, can help given they can return answers in milliseconds in many cases.
But the case for decision APIs at the edge isn’t just about latency. It’s about volume, Carrera said. In a world where decision APIs are embedded across consumer products, there will be too much traffic to send and process centrally.
“There’s a limitation on what you can serve from big data centers…That’s not going to work. You need distribution of that,” Carrera said. It’s the same principle as content distribution networks.
He expects traffic patterns to change as well, with more machine-to-machine communication. “You will have agents talking to other agents that may be doing transactions at the edge of the network, not because of latency, but proximity, reducing traffic, and having interactions between them,” he explained. That’s why this moment is different from the edge conversations that were happening before.
Chua isn’t sold yet though. He said that most use cases once pitched for the edge ended up on-premises, on the device or in a regional cloud. And because decision models are small, some could realistically run on a phone or other device rather than at the edge.
“In a base station or a central office, I don’t really need it,” he said.
Three reasons telcos should care about decision APIs
But that doesn’t mean telcos should write decision APIs off completely.
First and foremost, Chua said telcos should look at how they can use decision APIs internally for use cases that have constrained outcomes and need faster responses. He pointed to fraud detection and model routing as some of the ways telcos could put decision APIs to work.
Second, he said telcos can look into how decision APIs might enable them to provide new services for customers.
And finally, while he’s skeptical, he said they should also dig into what it would mean to host decision models for potential use cases around smart cities and physical AI.
“Just because of the hype and buzz around it, it is something that the telcos need to be aware of,” he concluded.
Read more about AI models on Fierce Network:
Open models are driving AT&T’s AI ‘tokenomics’ strategy
T-Mobile says open AI models aren’t just about cost — they’re about fit
Model skew is a hidden pitfall for telco AI, experts from AT&T, Boost and GSMA warn
The Five Nine: How AT&T and GSMA are building AI for telecom networks