Yesterday OpenAI unveiled a bunch of new tools that bring AI one step closer to doing tasks for you, not just answering questions. The big news was AgentKit (including a visual Agent Builder and ChatKit), plus ChatGPT Apps. In plain terms, this means ChatGPT can now integrate interactive “apps” (like Spotify or Booking.com) right in the chat, and developers get drag-and-drop tools to build smart agents. Let’s break it down in everyday language.

ChatGPT Apps: Doing Things with Chat

ChatGPT isn’t just for Q&A anymore – it’s becoming a platform that can do things. You can now summon apps right inside the chat. For example, you might type “Booking.com, find me a hotel in Paris” and ChatGPT will display actual hotel listings (with photos and prices) right in your conversation. The first set of apps includes big names like Booking.com, Canva, Coursera, Expedia, Figma, Spotify, and Zillow. These apps launched on Oct 6, 2025 (OpenAI’s Dev Day) and are available now to most ChatGPT users (outside the EU). (More partners like AllTrails, Peloton, Uber, and others will roll out later this year)

Because these apps run in the chat, you just converse normally. To use one, start a message with the app’s name and what you want. For example, say “Spotify, make a playlist for my party this Friday” and ChatGPT will automatically bring up Spotify’s interface to create that playlist. The first time you try an app, ChatGPT will ask to connect it (so you know what data is shared). ChatGPT can also suggest apps when relevant – e.g., if you’re talking about houses, it might pop up Zillow to help you browse home listings. In short, you get interactive mini-apps (maps, playlists, slide decks, etc.) woven into the conversation.

AgentKit & Agent Builder: Building Your Own AI “Helpers”

At the same event, OpenAI also launched AgentKit – a toolkit for building, deploying, and optimizing AI agents (automated workflows). The star is Agent Builder, a visual drag-and-drop canvas for designing complex AI workflows. Instead of coding every step, you drag blocks (like “Ask User”, “Search Web”, “Generate Text”) onto a canvas and connect them with arrows to define the logic. OpenAI’s CEO even called it like a “Canva” for agents – it’s a fast, visual way to design multi-step processes. For example, they showed teams who built sophisticated support or sales agents in hours instead of months by using this interface.

Agent Builder lets you compose multi-agent workflows with nodes and connectors. In the screenshot above (from OpenAI), each node represents a step or tool (like “Classification agent” or “Hallucination guardrail”) and you can wire them together. The interface even supports versioning and preview runs, so you can iterate quickly. In one demo, a company said it slashed an agent’s development time by 70%, getting it live in two sprints instead of two quarters.

AgentKit bundles more than just the visual builder. It includes ChatKit, an embeddable chat UI so you can drop a GPT-powered chat window into your own app. This means you don’t have to code a chat interface from scratch – ChatKit handles streaming responses, threading, theming and all the UI “plumbing”. (Companies like Canva and HubSpot have been able to integrate ChatKit in just minutes in demo setups.)

It also includes a Connector Registry (in beta) to manage integrations. This registry lists common data sources (Dropbox, Google Drive, SharePoint, Microsoft Teams, etc.) that your agents can access securely. Instead of writing custom API code for every tool, you can hook into these pre-built connectors. Finally, AgentKit boosts testing with new evaluation tools. You can define test sets, grade whole-agent outputs step-by-step, and even run automated prompt optimizations. All of these are designed to make agents more reliable and safe.

As for availability: AgentKit’s pieces come under OpenAI’s standard API pricing (you pay for model usage). On launch day (Oct 6), ChatKit and the new eval tools were generally available to all devs; Agent Builder went into public beta; and the Connector Registry started rolling out to certain enterprise and education customers.

Why It Matters (Pros and Cons)

These updates are exciting for several reasons:

• Built-in apps: ChatGPT now does more than chat – you can interact with real apps for booking hotels, making playlists, designing slides, and more, all in plain language. And with 800+ million ChatGPT users, app developers instantly get a huge audience
• Visual agent-building: Agent Builder democratizes AI automation. Even non-experts can sketch out an agent’s workflow visually. This could dramatically speed up creating complex assistants – OpenAI notes some companies build agents in hours that used to take weeks
• Plug-and-play chat UI: ChatKit means developers don’t have to reinvent the wheel for chat interfaces. A consistent, branded chat UI is ready to go, so your agent feels polished and integrated
• Enterprise features: The Connector Registry and eval tools bring structure for businesses. Agents can connect to common enterprise tools securely, and teams can systematically test and tune them (instead of “duct-taping” monitoring)

But there are some caveats:

• Early days & rollout: As of now, many features are in beta or limited regions. ChatGPT apps are live outside the EU (EU availability is promised soon). Agent Builder is still labeled “beta” and the connector panel is phasing into enterprise plans. So general users and smaller teams may have to wait or experiment cautiously
• Cost and complexity: These agents use OpenAI’s models under the hood, which means usage costs money (it’s all on metered API pricing). While the visual tools help, designing a robust multi-step agent still takes planning. For routine automations, it might be overkill
• Data and privacy: Connecting apps and agents means sharing data between services. OpenAI says apps must follow privacy rules and will prompt you on what’s shared, but organizations will need to review compliance. If an agent reads your emails or business data, you have to trust that flow
• Not a Zapier killer (for now): It’s tempting to think this replaces integration platforms like Zapier or n8n. Some commentators have even punned “Bye bye Zapier” in agent articles. In reality, they serve different needs. Traditional tools are great for fixed automations (“when this happens, do that”) with thousands of connectors, whereas AgentKit is about building intelligent, context-aware workflows. In one analogy: n8n is a switchboard of actions, while AgentKit is a “brain factory” where AI agents plan and decide on the next steps. So you’ll probably use both: Zapier/n8n for standard triggers, and AgentKit when you need something to think. Deterministic automation and smart generative AI automation are complementary and not alternative paradigms

(Nerd’s personal take:) I’m genuinely excited about these updates. It’s kind of sci-fi to be dragging boxes around and having an AI assistant ready in minutes. I love that ChatGPT is turning into an app platform – the idea of just asking for things in chat and getting them is awesome. On the other hand, I won’t be ripping out my Zapier automations overnight. Those reliable tools have their place. For now I plan to keep using them for simple tasks and save AgentKit for when I really want AI-savvy helpers. It’s a big leap in capability, but I’ll stay cautiously optimistic (nerd out, definitely 🙂).