Gagner grâce à des produits propulsés par l’IA : Conor Spicer, ingénieur solutions, OpenAI

Hi everyone, good afternoon. I'm Connor Spicer. I'm going to talk now around actually how we build new products with AI and how that can completely change our clients experiences. So I want to introduce you to Codeex. Codeex is our AI coding agent um that can do much more than just autocomplete. It can completely automate software development cycles and many different tasks that developers will be doing. We can give it a job, have it look across our codebase and run for many hours or perhaps even a full day until the work is done. So back in February, we launched the Codeex desktop app. It allows anyone with natural language to start generating code and completing many different tasks. We think this Codex app is actually now the ideal surface for both technical and semi-technical teams across financial services and adoption of Codex has basically broken all of our charts. When the app was launched back in February this year, it passed a million downloads in the first week. Um, and from there it's just kept on growing. We've now got over 4 million weekly active users uh across our codeex tools. And inside OpenAI, our engineers are using Codeex by default. We now ship more in a week than we were doing in a month previously. We're seeing 50% more PRs, that's chunks of code completed and submitted per engineer. So, we're massively able to increase the output of code that we produce, the products that we can ship without increasing headcount the same amount. It hasn't replaced our engineers at all. In fact, what we're seeing is it's changing their workflows and making them dramatically more effective in what they do. So now imagine this in your world. We can start refactoring and migrating legacy cobalt systems. We can automate regulatory reporting creating audit ready documentation or we can rapidly prototype across lending, trading or payments products and services which we offer. So this is the future of development. Um shortly I'll go and show a demo of what this looks like and how it works. Um but to walk through this we're going to use the scenario of Blossom Bank. Blossom is a successful bank with a really strong consumer banking product but are facing threats from competitors and growing pressure to move quickly to implement new features that their customers are demanding. One request that we keep seeing coming up from the customer base is around predictive budgeting. They don't just want to be told what they've spent anymore, but they want tools to help them anticipate their future spending and help them plan ahead with that. And actually, they're hearing this a lot. So, they are under pressure to move quickly with this. But this is the type of task that would have taken a long time before and a lot of coordination across different teams to work on this. So, that's why they're turning to Codeex. In this demo, I'll now show you how we can transform these workflows both inside the Codeex app to build our code and get our job done much more efficiently, but also within GitHub and how we can use this to review the code we produce and make sure that it's safe and secure to ship. So, with that, we'll jump into our demo. So here we are inside the codeex application and this is all powered by that very powerful GPT5 model which we have. Firstly let me pull up our consumer app. So this is what we want to work on. We have a running simulation here in our de development environment and keep an eye on this weekly spend. This is the feature which we want to change from being historical looking into that predictive forecasting tool that so many of our customers are asking for. As much as I might want to get in and immediately start writing the code, the reality of my job is I probably have other things to be doing. I may be pulled into an urgent meeting around some instance that we've had that I need to grab other data for. Codeex can help for this. I can ask codeex to look through the context it has of our codebase and our documentation and other tools to help understand this for me and quickly grab whatever it is as soon as I realize that I need to have it. We'll see it looking across the different observability tools and the code bases to pull the summary for me. This is the sort of thing that would have needed coordination across many other teams previously but now can be run immediately even live in a meeting if the topic comes up and I wasn't expecting it. Ad hoc tooling is great. We can ask when it we want it but we also want to scale this across our teams to take these best practices and to run them whenever I need and not be reliant on me deciding that it's a good idea to do this. What you can see here is some template solutions that we're seeing a lot of our teams adopting and using that are helpful to get started. But I can also go in and create my own automations around a weekly engineering summary where I need to understand what is it that we've built, what is it that we've shipped, potentially any issues that are blocking our team that I can helpfully have ready for me at the start of the week. So now that we've freed up time, I'm able to get back to what I enjoy doing, which is getting the code built, seeing these new features come to life. We can imagine that actually the context for this sits in many different areas. I may have a management team who've approved a certain definition of the scope that they've worked on with design and product teams and that's all set in SharePoint. Instead of me having to switch out across multiple apps, I can just ask codeex to use that connector and go and find it. Similarly, it could pull in context from tools like Jira, notion, even my emails if people had sent me updated specs in there. Once this starts running, codeex will understand my request to build out the plan of how are we going to implement this. We can see it here searching for SharePoint to find the file that I've referenced. And with that, I can start to see the stream of what it's working on. I aren't not having to click through manually and search for it, but I can still inspect how codeex is working. Now that it's got that file, I can see it's the one that I was expecting and it's going to start inspecting my codebase and immediately build a plan. I can open this up directly in the app. Again, avoiding switching across many different tools, understand the proposed approach, validate before it starts changing any code that this is what I expect. This looks good to me. So, I'm going to go ahead and say that we can begin to implement this plan, but also including in here some other tasks. Maybe I want to run tests to check that the code it writes is going to match anything, any standards that we have before it starts to run for me. So, I'll submit this and this is where codeex really gets to work. It starts inspecting my files and immediately starting to implement this feature across all of the front end and backend services which I'm running here. This is the type of task where engineering has completely changed. Instead of typing the code out myself, I'm prompting this. I'm inspecting how the agent is building it. I can steer it at this stage before it completes. Say actually I have a new idea. Um and once completed like this can go in and inspect actually what is the code that has been written. Having it providing me the context of this I can check and then build my own understanding of that code and see a summary of how it's all worked. Switching back to the app, we see the new feature is implemented. It's been prototyped. It's ready immediately to go and start collecting feedback on to share to other people. Finally, I can commit this and push all of this new code. So, it's ready for review. Now, that might seem really simple. Um, and I'm sure some of you are thinking, well, actually, we have regulators to deal with. We have friction with legacy portals and we have a lot of other challenges. Well, the good news is codeex can help here, too. Imagine I have this type of very legacy looking portal that may not have APIs to integrate with, but I need to submit all of the summary of my changes to consumerf facing apps to the regulator. This is something that would be a blocker to a fast pace of change and that would take a lot of time to pull the requirements together for it. Using a skill in codeex and our browser automation tooling, codeex can completely handle this task for me. understanding what the form requires, going in, searching across my code to find the relevant pieces of information, the documentation, and the evidence for this before completely populating that portal for me. So, at this point, it's found what it needs. At this stage, I'm hands off. I'll let the automation take over. it can go into the browser and start filling out all of these different pieces of information for me where it will validate it and save the draft. But keeping me in the loop, it won't just hit submit. It will pull together a summary of the changes which it's made, share that with me and allow me to then check are there any final points that I want to implement or do I simply want to read through this, check it myself and then hit submit. Again, taking tasks that would have taken hours previously down to something which can run in a couple minutes or less. So, we're increasing the amount of code that we're shipping. The other aspect we want to do is how do we review that and make sure that it's safe as easily as possible. Here we are in GitHub where I have my new change committed. Our automated set of tests have run and our human reviewers have checked this and approved this. But we can also have codeex embedded in this process where it automates the review process. And in this case, we see that it's actually found a cyber issue, a potential mishandling of sensitive fields that was missed by our human reviewers, but has been caught here that we can then even send to codeex to start implementing the fix for us. So I want to leave you um at this point with a couple key takeaways which we have. We've seen how Codeex can really accelerate our development workflows and the other types of tasks that our developers and engineers are doing. And secondly, Codeex is able to not just ship more code, but to secure it through a production ready contextual code review. And together, this combination of speed and safety is driving so much excitement, which we're seeing around codeex today. Let's be honest with ourselves though, we know there strain on organization with all this new code, new tooling to learn. And that's why myself and our team focus so much on enabling and advising our customers engineering teams around how we scaffold these new processes so that when we scale up the volume of code we can really meet the moment and create this overall win for our customers.