Good Morning! OpenAI just dropped GPT-5.5, a massive step toward their highly anticipated "agentic superapp," built using recursive self-improvement. Plus, i’ll show you how to Face Swap using ChatGPT's latest image model for free.
OpenAI Launches GPT-5.5 for Agentic Workflows
Meta Slashes 8,000 Jobs to Fuel AI Push
Claude Managed Agents Get Built-In Memory
How to Face Swap Using ChatGPT
4 new AI tools worth trying
AI MODELS
Just seven weeks after launching GPT-5.4, OpenAI took the wraps off its latest flagship model, GPT-5.5. The team revealed that the model itself and OpenAI's agentic coding tool Codex were used to build it, marking a dramatic leap forward through recursive self-improvement (AI building better AI).
Acts as the core engine for OpenAI's upcoming "superapp," which aims to transition Codex from a pure developer tool to a consumer automation app for anyone working on a computer.
Vastly improved at long-horizon tasks that require multi-step planning, intuitive problem-solving, feedback loops, and tool use.
Features upgraded "computer use" capabilities, with OpenAI claiming it can now perform economically valuable work autonomously across 44 different occupations.
Handles complex scientific workflows, including generating novel mathematical insights and analyzing messy real-world data.
Reports zero impact on latency despite massive capability upgrades, directly addressing the lag issues currently seen in rival frontier models.
Ships with upgraded safety protocols, including third-party bioweapon and cybersecurity red-teaming, resulting in stricter refusals for misuse attempts.
We are officially witnessing recursive self-improvement in a commercial product. By optimizing GPT-5.5 strictly for agentic workflows and computer use without sacrificing speed, OpenAI is transforming AI from a passive chatbot into an active "orchestrator." The ultimate goal is no longer just answering questions, it's shifting to a fully compute-powered economy where AI handles the heavy lifting of your daily workflows.
AI NEWS
Meta CEO Mark Zuckerberg has announced a massive 10% reduction in the company's global workforce, cutting roughly 8,000 employees and scrapping 6,000 open roles as the tech giant pivots aggressively toward generative AI.
The layoffs, starting May 20, follow previous targeted cuts in Reality Labs and the recent replacement of third-party human content moderators with AI systems.
To train future AI agents, Meta is actively logging employee keystrokes and mouse clicks using a new internal tracking tool called the "Model Capability Initiative."
This is part of a brutal industry-wide trend: Microsoft just offered voluntary buyouts to 7% of its US staff, and Amazon recently cut another 16,000 corporate roles to reckon with the AI boom.
The cuts come as Meta ramps up its AI investments, having recently debuted its first major model since aggressively hiring Scale AI's Alexandr Wang.
This directly echoes the mathematical "AI Layoff Trap" researchers just warned about. Tech giants are systematically cannibalizing their human workforce to fund an incredibly expensive arms race in autonomous agents. When companies are literally harvesting their remaining employees' keystrokes to train the very AI systems that will eventually replace them, it signals a ruthless, permanent restructuring of the tech labor market.
AI TOOLS
Anthropic just launched filesystem-based memory for Claude Managed Agents in public beta. Instead of starting from scratch every time, your agents can now learn from every session and share insights with each other, radically improving their long-term performance.
Memory mounts directly onto a filesystem, allowing Claude to seamlessly use its existing bash and code execution capabilities to read, write, and organize its own context.
The system is built for enterprise deployment with scoped permissions, meaning you can have read-only org-wide memory stores alongside read-and-write per-user stores.
Memories are stored as files that developers can export, redact, or manage via the API, complete with full audit logs to trace exactly what an agent learned and when.
Early enterprise testing shows massive efficiency gains: Rakuten cut first-pass errors by 97%, while Wisedocs sped up document verification by 30% simply by letting agents remember recurring issues.
Cross-session memory has historically been one of the biggest bottlenecks for scaling agentic AI. By providing an observable, filesystem-based memory layer, Anthropic is removing the need for developers to build their own clunky retrieval infrastructure. Now, autonomous agents can actually compound their knowledge over time, turning AI from a stateless tool into a continuously learning employee.
HOW TO AI
🗂️ How to Face Swap Using ChatGPT
In this tutorial, you will learn how to use ChatGPT’s latest image model, GPT Image 2, to seamlessly swap a face from one image onto a target body. This new architecture is significantly more consistent with lighting and shadows, making professional-grade edits accessible through simple conversation.
🧰 Who is This For
Content creators making thumbnails, posts, visuals fast
Designers creating assets without starting from scratch
Marketers generating ad creatives and campaigns
Social media managers scaling daily content
STEP 1: Upload Your Images in the Right Order
The order in which you upload your files is crucial for the AI to understand your prompt logic. Open a new chat in ChatGPT and upload the target body image first. This is the photo that provides the background, clothing, and posture. Immediately after, upload the source face image. By keeping this order, you can refer to them clearly as "Image 1" (Body) and "Image 2" (Face) in your instructions.
STEP 2: Use a Precision "Blend" Prompt
Once both images are uploaded, enter a prompt that clearly defines the source and target. To get the most realistic result, you must instruct the AI to respect the environmental conditions of the original photo.
Prompt: "I want you to take the face of the person from Image 2 and place it on the head of the person from Image 1. Ensure the final output matches the lighting, shadows, and skin tone of Image 1 to make the blend look natural. Use the same aspect ratio as Image 1."
STEP 3: Match the Facial Expression
A common beginner mistake is swapping a smiling face onto a serious body, which often looks "off." If your source face (Image 2) is looking at the camera but your target body (Image 1) is looking at a phone, you need to tell ChatGPT to adjust the gaze.
The Fix: Add this extra instruction: "Additionally, adjust the expression and eye direction of the face from Image 2 to match the expression of Image 1, ensuring the person is looking in the same direction."
STEP 4: Handle Group Photos with Descriptions
If you are swapping a face into a group selfie, you cannot just say "Image 1." You must describe the specific target person.
The Workflow: Upload the group photo (Image 1) and the face (Image 2).
The Prompt: "Take the face from Image 2 (the guy with the silly expression) and place it on the head of the guy on the far right in Image 1 who is wearing a blue shirt. Keep everyone else in the photo exactly the same." This "Identity Lock" capability ensures the rest of the group remains unchanged.
Anthropic says it has fixed three causes of recent Claude Code quality issues: reduced default reasoning, a caching bug, and a system prompt to reduce verbosity.
Instagram launches Instants, an app for sharing disappearing photos, in Italy and Spain, after rolling out an Instants feature in its main app in some regions.
Microsoft says Copilot's agentic features in Word, Excel, and PowerPoint are now generally available and enabled by default for 365 Copilot and 365 Premium.
Vercel says some customer accounts were compromised prior to its early-April breach, potentially through social engineering, malware, or other methods.
🎆 ChatGPT Images 2.0: OpenAI’s upgraded image generator
🌎 GPT5.5: OpenAI’s most powerful AI model built using recursive self-improvement.
🌎 Tiny Aya: Cohere's small, open-source model covering 70+ languages
💻 Holo 3: Open AI agent that can use computers like a human
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas