ChatGPT reçoit sa plus grande mise à jour de tous les temps

Chat GPT is about to change more than it has at any point since 2022. And OpenAI's goal is obvious. Turn it into the main AI layer for your digital life. And the timing says everything. Codeex is exploding. Anthropic is attacking the developer market. Open AAI is trying to lock down its own chip future. And Chat GPT is becoming the front door to the whole strategy. According to several reports, OpenAI is getting ready to roll out a major overhaul of chat GPT over the coming weeks. And the goal is to turn Chat GPT from a place where people ask questions into something much closer to a full AI super app with coding tools, image generation, thirdparty apps, workplace automation, personal scheduling, enterprise services, and agents that complete tasks for users. One of their senior employees literally said, "The era of using AI just for chatting is over." Which is a pretty bold statement coming from the company that basically made AI chat bots mainstream. Tibo Satio, the OpenAI executive leading core products and platforms, described the future version as something people can connect to through their phone, desktop, or web and even talk to while they are in the car. The idea is that chat GPT becomes a personal agent across life and work. Instead of opening a chatbot, typing a prompt, getting an answer, and doing the real work yourself, the long-term vision is that you give the agent a task, and it figures out which tools to use and how to finish it. And the most important part of this overhaul is Codeex. Codeex is OpenAI's programming tool, and it is one of the fastest growing products inside the company right now. OpenAI is preparing to integrate codecs much more deeply into chat GPT, turning the normal interface into a gateway for software control, coding tasks, workflow automation, and general computer work. Alex Emiros, OpenAI's head of enterprise products, described this as the second stage. The company already has an intelligent agent that can help users do things on a computer. The next goal is to bring that capability to everyone, including people who do not think of themselves as developers. OpenAI has reportedly launched six Codeex plugins for job roles like creative production, sales, and public stock investment. So, Codeex can move beyond programmers and become useful for office work, business tasks, and personal productivity. For normal users, the first visible changes will probably show up inside the chat GPT website and mobile app. OpenAI is expected to add more prompts, buttons, and entry points that guide users toward coding tools, image generation, and thirdparty applications from partners like Canva and Booking.com. In the early stage, users may even get the option to manually choose whether codecs or regular chat GPT should handle a request. The examples being discussed are very practical. After connecting email or Slack, a user could dictate a request and Codeex could draft the message, choose the recipient, prepare everything, and send it after confirmation. After connecting a calendar plug-in, someone could ask about upcoming events, or quickly pull up meeting background. A user could even set a workflow that runs every morning at 8, checks the day's meetings, scans the inbox, and pushes the most urgent tasks. That is the real shift from answer my question to handle this task. Satio's longerterm vision goes even further. He believes the transition period with visible buttons, prompts, and tool selection may not last very long. Eventually, the model should automatically understand what the user wants and decide whether the task should run locally in the cloud, through codecs, through a plug-in, or through some other tool. The user should not have to care which tool is doing the work. That is why some people online saw this as an early step toward AGI style product design where the interface slowly disappears and the agent becomes the real product. The technical reason OpenAI believes this is possible comes from model upgrades. GPT 5.5 released in April this year is much better at long-term multi-step tasks than earlier models. It needs less manual guidance, giving Codeex more confidence to execute work after clear instructions. Now, you've probably noticed how much attention Claude is getting right now. Anthropic keeps adding new models and features from Claude code and Claude artifacts to skills, connectors, design tools, and more. And honestly, it makes sense. Claude has become one of the most useful AI tools for turning an idea into something real. Whether that means building an app, creating a presentation, organizing research, planning your week, or speeding up work that would normally take a whole team. The problem is that a lot of people keep saying learn claude without actually showing you a clear way to use it properly. That's why today's sponsor is hosting the world's first Claude Aathon, a 2-day live workshop happening this weekend from 10:00 a.m. to 7:00 p.m. Eastern time. It's a deep dive into Claude practical use cases and more than 10 other AI tools. And they're opening 1,000 free seats for a limited time. Inside the workshop, you'll learn how to use Claude for deep research, build artifacts and dashboards, create full presentations, set up connectors like Indeed for job search, build custom GPTs and agents, and use AI tools for visuals, videos, and automation. You'll also get bonus resources like claude codes, a prompt library, and a personalized AI toolkit builder. Link is in the description or scan the QR code to join before the free seats close. All right, now back to the video. Mitch Trojanowski, co-founder of Basis, reportedly said that by GPT 5.5, Codeex's advantages became too significant to ignore. That point is important because coding agents need trust. Developers are careful about giving AI control over real code. Yet more of them are becoming comfortable handing a clear plan to codeex and letting it complete the work from start to finish. Verun Ralph, an engineer at Notion, said that while he and his colleagues were building a tool, Codex had been silently fixing a bug every time. They only realized the bug had existed for weeks when they tried other agents that did not fix it automatically. That kind of experience makes the tool feel like part of the workflow instead of a novelty. OpenAI also lowered the barrier by launching a Codeex desktop app in February. This made sense because Claude Code, Anthropic's competing coding tool, was originally used through a pure text command line terminal. That is fine for experienced developers. Yet, it creates a big wall for regular users. A desktop app gives Codeex a more familiar graphical interface while still helping power users who run several agents at the same time. Max Showning, a product manager at Notion, reportedly likes starting a codeex task on a computer and checking progress later on the mobile app, which is exactly the background agent behavior OpenAI wants to normalize. And the numbers behind Codeex explain why OpenAI is pushing so aggressively. In midMay, Google search interest for OpenAI's codeex reportedly hit a record high and passed clawed code. Since the desktop app launched, Codex's user base has grown sixfold in less than two months. Weekly active users reportedly passed 5 million at the end of May. Sam Alman said Codex usage is rising 5% per day and Greg Brockman reportedly said enterprise codeex revenue increased 50% week-over-week. OpenAI has almost 1 billion chat GPT users or more than 900 million consumer level users depending on the report. Yet most of them still use it for free. Culturally, that is a massive win. Financially, it leaves OpenAI with a huge monetization problem. Chat GPT became one of the biggest software products in history. Yet, the company now needs to prove to investors that this giant user base can turn into serious recurring revenue. That is why the conversation keeps coming back to enterprise customers. Around 2 million businesses use OpenAI products, and they reportedly account for around 40% of OpenAI's revenue. The company expects that share to rise to 50% by the end of this year. Codeex fits that plan because coding tools and productivity agents are easier to sell to companies than casual chat features. This also explains some of OpenAI's recent product decisions. In January 2025, OpenAI launched Operator, an agent designed to use a web browser like a human, doing things such as booking tickets or shopping online. The idea sounded powerful, yet management reportedly realized that making AI imitate human clicks on websites is inefficient. Sio said directly, "Writing code is a more effective way for AI to handle tasks like building complex spreadsheets on a computer." If code is the better interface for getting work done, then codeex becomes more than a developer tool. It becomes the foundation for a general AI agent. And this is where Anthropic enters the story. OpenAI's move was pushed, at least partly, by competition from Anthropic. The rivalry became more serious in autumn 2024 when internal benchmarks suggested that Anthropic's models had moved ahead of OpenAI in programming ability. Then in February 2025, Anthropic released the preview version of Claude Code, and OpenAI realized a smaller competitor had overtaken it in one of the most important AI categories. For open AI, losing encoding is a huge problem because coding is tied directly to its belief about AI progress. If AI can write better code, it can speed up AI research, infrastructure, product development, safety work, and eventually the race towards super intelligence. Falling behind in programming means falling behind on the track that may matter most. Jenny Shaw, a former Open AAI researcher and now a partner at Leona's Capital, described the shift in a very interesting way. Around a year ago, OpenAI was focused on going all out toward the bigger dream, while Anthropic focused more on making money first. Now, the two paths are converging because both companies are moving toward IPO style expectations and investors care very directly about revenue. OpenAI responded by creating a dedicated codec research and development team led by Sdio giving it high autonomy and even open sourcing the code behind codecs to get direct user feedback. Then the structure changed. In January, OpenAI reorganized product teams around model researchers. In May, it merged the chat GPT codeex and API teams into one unified core product and platform department under SATO. That kind of restructuring usually means resources are moving away from certain projects. Some consumer-facing products became casualties. The chat GPT shopping checkout feature was put on hold and Sora, the video generation product launched less than a year earlier, was reportedly shut down. The strategy is pretty obvious. Move resources toward enterprise tools, coding agents, and services people are willing to pay for. Meanwhile, Anthropic is pushing hard from the other side. Anthropic's annualized revenue reportedly rose to 47 billion in May, driven by Claude Code and Claude Co-work attracting developers and enterprises. OpenAI's annualized revenue has reportedly passed $30 billion after previously announcing $25 billion in March. Even if the exact numbers need to be treated carefully, the direction is clear. The AI labs are fighting over revenue, developers, enterprise adoption, and the agent layer at the same time. On June 7th, reports said Clive Chan, described as employee number 002 on OpenAI's self-developed chip project, announced on X that he had left OpenAI and joined Anthropic. Chan spent two years at OpenAI, said he was the second person recruited into the hardware team, watched the chip project grow from its early formation, and praised the team as one of the strongest chip design teams in the world. Still, he said he could not shake the desire to climb a new mountain from the bottom again. So he chose Anthropic. He said Anthropic's talent, values, and ambition impressed him deeply, felt the work rhythm was intense after only a few days, and summed it up with one line, "It's time to build." When people asked him about OpenAI's chip project, he said he could not reveal more beyond what had already been disclosed publicly. He pointed to the OpenAI and Broadcom cooperation blog from October 2025 where the two companies described a 10 gawatt OpenAI designed AI accelerator system. OpenAI is responsible for chip and system design while Broadcom handles accelerator and networking development and deployment. The first racks are expected in the second half of 2026 with the full project planned to run until the end of 2029. Chan's background also shows why this move caught attention. He graduated from the University of Wateroo in 2021, helped build organizations connected to Canadian Hyperloop, and worked or interned at Google, SpaceX, Quer, Tesla, and OpenAI. His work covered machine learning infrastructure, liquid rocket engine projects, GPU optimization, cluster scheduling, data center software, training infrastructure, matrix multiplication, roofline analysis, hardware performance, and OpenAI's AI chip effort. His move to Anthropic became another symbol of how intense the rivalry has become. Anthropic employees welcomed him publicly, while people online joked that every I have decided to leave OpenAI post now seems to end with Anthropic. Others compared it to leaving Real Madrid for Barcelona, while some question the starting from the bottom line since Anthropic's valuation is reportedly close to $1 trillion after a massive series age financing. And while OpenAI is trying to turn Chat GPT into a more powerful agent platform, it is also dealing with the risk that comes with giving AI more access to the outside world. That is where lockdown mode comes in. OpenAI recently announced a feature called lockdown mode designed for users and organizations handling sensitive data. The point is to reduce prompt injection risks, one of the biggest security problems for AI agents. A prompt injection attack happens when malicious instructions are hidden inside content the AI reads such as a web page document message or external data source. If the AI follows those instructions, it could leak data, misuse tools, or behave in ways the user never intended. Lockdown mode is basically a restricted version of chat GPT for sensitive work. It disables web browsing, image display and responses, deep research, agent functions, networking with canvas code generation, and file downloads. OpenAI says the mode is not intended for everyone. It is designed for people and organizations that want stricter protection from data exfiltration risks related to prompt injection. So, this new chat GPT overhaul is really about three things happening at once. OpenAI is trying to turn its massive free user base into paid productivity demand, using Codeex as the engine behind that shift and racing anthropic in a market where coding agents, enterprise revenue, AI chips, and top technical talent are becoming part of the same war. The old chat box made Chat GPT famous. The next version is being built to make it useful enough that people and companies actually pay for it. Also, if you want more content around science, space, and advanced tech, we've launched a separate channel for that. Links in the description. Go check it out. Hit subscribe if this made you rethink just how fast all of this is really moving. Thanks for watching and I'll see you in the next one.