I’ve been using GPT-5 for more long-running, complex tasks. For example, using it with Background Agents for independent tasks and parallel foreground agents when switching between different tasks and reviewing their outputs.

GPT-5 is one of the most steerable models I’ve used. I’ve needed to be more explicit about what I was trying to accomplish for some tasks. Leaving things vague ended up with the model taking a different direction than I expected. When I was more specific, I was surprised by how smart the model was.

There are a few places where I’ve gotten stuck in the past week coding where GPT-5 solved a complex bug other models couldn’t figure out. I’ve used it for tasks like looking through our Stripe queries and figuring out the best optimizations to remove latency.

The default writing style of GPT-5 was more verbose than I wanted, so I was able to make a rule to have responses be more concise. I also prefer more agentic behavior, so I was able to steer the model towards asking fewer follow-up questions after prompting. I’ve been using it for more difficult coding problems.

My favorite way to taste test is to use the same prompt for many different models and see if any can one-shot what I was expecting. Our frontend and backend use protobufs, which sometimes throw models off. I asked GPT-5 to build a backend API endpoint as well as the corresponding frontend React component, which are in two separate submodules in our monorepo. It was able to get this code correct and regenerate the protobufs to fix type errors in one shot.