Sam Altman says AGI is coming this year

Hola Decoder😎

If someone forwarded this to you and you want to Decode the power of AI and be limitless, then subscribe now and Join Decode alongside 30k+ code-breakers untangling AI.

🧠 Sam Altman says OpenAI could reach AGI this year

Sam Altman has put one of his clearest timelines yet on AGI, saying OpenAI expects to have an internal model by the end of 2026 that meets his personal definition of artificial general intelligence. 

The Decode:

OpenAI says it is 80% of the way there - Research chief Mark Chen estimates the lab has covered roughly 80% of the path to AGI, while Greg Brockman believes this period may eventually be remembered as when AGI arrived.

Astra can reportedly do a week of research alone - Chief scientist Jakub Pachocki says the model can take a research paper and independently perform roughly a week of researcher-level work, meeting OpenAI’s 2026 “automated intern” target.

Altman says Astra can create new knowledge - He described Astra as OpenAI’s first model capable of “actually inventing new things in a way that matters,” arguing that meaningful knowledge creation is a strongly AGI-like capability.

OpenAI is also confronting the risks - Following the security breach detailed in July, the company says it is putting greater emphasis on safety and slowing aspects of frontier development as capabilities continue advancing.

The important distinction is that Altman is talking about an internal model meeting his own AGI threshold, not declaring that universally agreed AGI arrives this year. But if Astra can independently conduct days of useful research and generate genuinely novel discoveries, the argument over whether AGI has arrived may increasingly become one of definitions rather than capabilities.

Together with AirOps

The Growth Targets Changed. Did Your Plan?

Growth targets are rising faster than the resources behind them, forcing marketing teams to make critical budget decisions without a credible view of how peers are responding.

AirOps surveyed 300+ CMOs and VPs to reveal how leaders are reallocating budgets and closing capability gaps as pressure grows to deliver more pipeline with limited support.

• Compare how peers are responding to higher targets, now affecting 75.4% of marketing leaders across the market.
• Build a stronger budget case using the reality that only 43.0% received more funding.
• See why 86.6% prioritize AI search while just 23% feel confident measuring its impact.

The report lets you compare your growth targets and available resources against responses gathered directly from 300+ marketing leaders.

You’ll immediately see whether your performance gap starts with execution or a plan that was under-resourced from day one.

Get the Marketing Leaders Reality Index.

🥸How to turn any meeting into a structured action plan using Claude

Most meetings end with vague notes nobody acts on. Here's how to fix that.

Step 1: Paste your meeting transcript or notes - Drop your raw meeting notes into Claude, messy, unstructured, whatever you have. Voice memo transcripts work too.

Step 2: Run the extraction prompt - "Extract every decision made, every action item assigned, every open question unresolved, and every blocker mentioned. Format it as a clean action plan with owner, deadline, and priority for each item."

Step 3: Generate the follow-up email - "Write a follow-up email to all attendees summarizing decisions, action items with owners and deadlines, and next meeting agenda. Tone: direct and professional."

Step 4: Build a skill for recurring meetings - If this is a weekly standup or recurring sync, save the format as a Claude Skill, so every future meeting outputs the same structure automatically.

🧠 AI Assistants Are Starting to Remember and Act

OpenAI and Anthropic are pushing their assistants beyond answering questions. New updates give ChatGPT more ability to act across logged-in websites, while Claude can carry what it knows about you between different workspaces.

🧠 Claude Now Remembers Across Chat and Cowork

Claude’s memory now carries across regular chats and Claude Cowork, so a Cowork task can start with context about previous projects, preferences, goals, or clients.

Users can see, edit, and delete individual memory topics in Settings, explicitly ask Claude to remember something, or disable memory entirely. Sensitive topics remain excluded unless enabled. You can see how it works.  

🔐 ChatGPT Work Can Now Log In for YouChatGPT Work can use password autofill to sign into frequently used websites while keeping the actual login credentials private. You can watch the release here.

OpenAI says this lets Work handle tasks such as booking appointments, completing applications, and other browser-based errands that require authenticated access.

AI assistants are becoming more useful between prompts, not just during them. Password access permits agents to act inside the services people already use, while persistent memory gives them the context to understand what needs doing without rebuilding that context every session.

Together with Playbook Pro

The Playbook for Profit That Funds Itself

Great operators don’t just count cash, they make it multiply. They use timing and leverage to turn everyday revenue into rapid expansion.

Financial Alchemy is your cheatsheet packed with real operator insights and ready-to-use templates. Inside, you’ll learn how to:

💰 Master cash-conversion cycles to grow while suppliers fund your inventory, just as Wayfair scaled with a –34-day cash conversion cycle.

📊 Use the Cash Flow Simulator to forecast scenarios and uncover up to $100K in hidden operating cash within a single quarter.

🧮 Pinpoint profit levers with the Unit Economics Analyzer to cut CAC payback windows from 6 months to as little as 3 months.

🤝 Negotiate better terms using the Supplier Toolkit, a framework inspired by Skinny Pasta’s strategy that extended supplier terms to 120 days and fueled 100%+ Y-o-Y growth.

Start compounding capital instead of chasing outside funding.

Get your ebook now and turn your cash flow into the most powerful growth engine!

🏆 Tools you Cannot Miss:

🎬 Screenify Studio – Records polished product demos on Mac using an AI agent, helping teams create professional walkthroughs without manually performing every click.

🧠 DeployHermes – Deploys persistent AI agents with dedicated roles, long-term memory, and reusable skills so they can keep working beyond a single session.

🔌 Knack MCP Server – Gives AI builders a HIPAA-compliant backend that agents can connect to through MCP for secure healthcare and sensitive-data workflows.

📹 Tellie Prompter 1.5 – Uses AI to track what you’ve already said while presenting and automatically keeps your teleprompter aligned with what comes next.

📊 Evidence Core – Lets coding agents build live analytics experiences directly from data, making dashboards and reporting more programmable and AI-native.

🚀 Quick Hits

💰 Nvidia is closing in on becoming a $100 billion-a-quarter company after posting record revenue of $96.2 billion. Data center revenue reached $89 billion, while Nvidia is forecasting roughly $108 billion in revenue next quarter.

🔥 Amazon is tripling its planned Nvidia GPU expansion after demand exceeded expectations. AWS will add another 2 million GPUs on top of the 1 million already planned and is also bringing Nvidia’s Vera CPUs to its cloud infrastructure.

🎙️ Google has launched Gemini 3.5 Transcribe, which automatically cleans up transcripts while supporting more than 85 languages. It can remove filler words like “um” and “uh,” recognize specialized jargon, format text and identify up to three speakers.

🤖 Bill Gates is proposing a tax on robots as governments prepare for AI-driven job displacement. He also wants some roles designated “Human Reserved,” arguing the measures could slow workforce disruption while generating money for retraining and stronger safety nets.

⚡ Anthropic has signed a $45 billion compute deal with Nscale, adding another massive infrastructure commitment to its AI expansion. The six-year agreement will give Anthropic access to compute powered by Nvidia’s Vera Rubin systems as it races rivals for capacity.

🤳Prompt of the day

Create a Cinematic Formula 1 Victory Portrait

GPT Image 2 Prompt: Create an ultra-realistic cinematic portrait of [INSERT PERSON], a [INSERT TEAM] Formula 1 driver, celebrating a race victory on the podium.

Context

• Driver: [Insert person / upload reference image]
• Team: [e.g., Ferrari / Mercedes / McLaren]
• Hair/appearance: [Insert details]
• Racing suit: Official team colors and authentic sponsor-style details
• Moment: Podium victory celebration
• Aspect ratio: Vertical 9:16

Instructions

Show [INSERT PERSON] celebrating with genuine happiness and adrenaline, wearing a realistic [INSERT TEAM] racing suit and matching podium cap while spraying champagne into the air. Capture sparkling champagne droplets frozen mid-flight with subtle natural motion blur, realistic liquid physics, and strong daylight reflections. Frame the driver from approximately the waist up at a slightly diagonal camera angle to emphasize movement. Use warm highlights across the face and hair, realistic skin texture, detailed fabric and sponsor patches, and a softly blurred Formula 1 podium in the background with cinematic bokeh. Preserve the subject's identity accurately if a reference image is provided, while keeping the pose spontaneous and physically believable.

🤳AI Nugget of the Day

Thanks for Decoding with us🥳

Your feedback is the key to our code! Help us elevate your Decode experience by hitting reply and sharing your input on our content and style.

Keep deciphering the AI enigma, and we'll be back with more coded mysteries unraveled just for you!