Why OpenAI's new AI agent tools could change how you code

monsitj/Getty Images

If you've been a production developer for any time, you know application programming interfaces (APIs) are subject to change at the provider's whim.

You might go to bed one night knowing your app is solid and has a bright future. Then you wake up in the morning to discover that one of your API providers has announced a change that requires you to spend the next six months recoding your app to work with the revised API. Ask me how I know.

Also: The best AI for coding in 2025 (and what not to use)

AI-based apps are not immune to this immutable law of API life. OpenAI has announced the introduction of the Responses API, which will trigger the sunsetting of its Assistants API sometime in 2026.

As with most API changes, the new API does more, works better, and has fewer calories. However, despite these capabilities, the new API still adds big to-do items to early adopters of previous APIs.

OpenAI has announced a series of new programmer-level features for building AI agents, including the API. Before we dig into what's new, let's put this change into context.

You've probably noticed that since generative AI (Gen AI) burst onto the scene a few years ago, nearly every product and service seems to have suddenly added an AI upsell feature to their base offering. It's a great line extension to many products and services, but it's not like each vendor had AI research teams building large language models.

Instead, these companies have licensed APIs that provide AI services at a fairly predictable fee per transaction. Different models cost different amounts, but at a rough estimate, a developer could make 67 GPT-4.5 typical chat queries, 1,176 GPT-4o queries, or 20,000 GPT-4o mini queries for a single dollar.

Also: How to use ChatGPT to write code - and my favorite trick to debug what it generates

With OpenAI's APIs (and other AI companies' APIs), AI functionality can be added to code with simple function calls and a credit card, cutting short years of development and millions of dollars of infrastructure and research costs.

The new Responses API

Until now, OpenAI offered an API, the Chat Completions API, for sending chat prompt queries to ChatGPT and getting a response. OpenAI also offered an API, the Assistants API, that enables agents to perform multi-step reasoning and access files.

The new Responses API combines the features of both previous APIs. While the Chat Completions API is expected to remain, the company has announced plans to sunset the Assistants API as soon as next year in favor of the more capable Responses API.

Also: How to turn ChatGPT into your AI coding power tool - and double your output

Rates for using Responses API are located on OpenAI's pricing page and vary based on model and usage.

The Responses API has three built-in tools programmers can call: web search, file search, and computer usage.

The web search tool

If you ever used ChatGPT to search the web and aggregate an answer, you know what the web search tool does. In the case of the Responses API, it allows programmers to initiate such queries from within their programs.

When used, the API provides text as a response, and links to sources and inline citations. Many of these responses should be correct. The company cites a chart showing that tests of web searches with GPT-4o were only 38% right (keep that in mind next time you use ChatGPT to search), but searches using the GPT-4o search preview and the GPT-4o mini search preview were correct nine out of ten times.

Mostly good is pretty good, right?

The file search tool

Next is the file search tool, which will search a library of your corporate documents. And yeah, I know what you're thinking. OpenAI says it doesn't train its AIs on business data. So, you should be fairly confident that uploading all your data will be secure and won't show up in someone else's response to a prompt.

Also: How I test an AI chatbot's coding ability - and you can, too

There are issues, though. For one thing, at $0.10 per gig, per month, file storage isn't cheap. My little company's corporate documents database is 57TB. If that database was uploaded to OpenAI for context-specific AI search, the storage would cost almost $6,000 per month -- and my company has just two employees.

The benefits of having an AI search tool is the ability to break down silos. But if the silos are what makes storing the data more affordable by only uploading select data, this pricing might limit the usefulness of this feature.

Computer usage tool

Finally, OpenAI includes its computer usage tool, the screen-viewing and clicking tool at the heart of its Operator agent offering.

This thing is supposed to be able to read screens and, based on screen content, either perform processing or click to cause actions. From the demos, especially in the preview of the current announcement, it looks like the tool has a long way to go before it's universally useful.

Stay tuned on this feature.

Agents SDK

An API is a set of endpoints or function calls for interacting with an external system. An SDK (software development kit) is a broader collection, often including software tools, some APIs, libraries, and utilities, that all work together.

OpenAI announced Agents SDK, a set of tools for building, debugging, and monitoring agents in action. The kit includes the Responses API and adds workflow orchestration, allowing coders to perform multi-step reasoning and task execution.

Also: This new AI benchmark measures how much models lie

The kit also allows coders to create teams of agents, each tasked with specific responsibilities. A "hand-off" functionality allows one agent to do some work, then hand-off processing to another agent, and so on.

Additionally, and I think this is one of the most important features, a monitoring dashboard allows developers to see each interaction with the AI, identify what agents were used for what, and how they were tasked. With the prevalence of AI hallucinations, it's important to be able to keep track of what all those independent tasks are up to.

What do you think about OpenAI's latest updates?

Are you planning to integrate the Responses API into your projects, or are you concerned about OpenAI's sunsetting of the Assistants API?

How do you feel about the built-in tools, such as web search, file search, and computer usage? Do they solve real problems or introduce new challenges?

And what about the Agents SDK? Do you see it as a game-changer for AI agent development, or are some features still lacking? Let us know in the comments below.

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.