Google’s Agent reaches Human Levels

Hello there! Ready to dive into another upgrading, Mind-boggling, and Value-filled Shift?

Today we have:

🎮 SIMA 2 Steps Into AGI Territory

🤖 Automate Sales, Marketing, and Ops in Minutes with Lindy

💸 OpenAI’s Leaked Financials Reveal a Bigger Cost Problem

🔨Tools and Shifts you cannot miss

🎮 SIMA 2 Steps Into AGI Territory

Google DeepMind unveiled SIMA 2, a Gemini-powered agent that can reason, act, and learn inside 3D virtual worlds and outperform its predecessor by a massive margin. It now follows goals instead of just instructions and adapts to unfamiliar environments with near-human-like flexibility. 

The Shift:

1. A reasoning agent that understands goals - SIMA 2 uses Gemini to interpret tasks, explain its plan, and act through a virtual keyboard and mouse, which shifts it from command-following to real-time collaboration.

2. Strong generalization across unseen games - It completes 45–75 percent of tasks in new worlds like MineDojo and ASKA, versus SIMA 1’s 15–30 percent, while handling sketches in multiple languages and even emojis with consistent accuracy.

3. Adapts inside fully AI-generated 3D worlds - Paired with Genie 3, it orients itself in new environments, executes user goals, and behaves sensibly despite having zero prior exposure to those worlds.  

4. Learns new skills through self-improvement - SIMA 2 trains through trial-and-error using Gemini to set tasks and score outcomes, letting it grow its abilities without new human demonstrations. 

SIMA 2 brings agents closer to general intelligence by showing they can reason, adapt and learn across limitless environments. This is the strongest proof yet that open-ended agents can teach themselves to reach human-level competence.

^{TOGETHER WITH RIPPLING}

Don’t get SaaD. Get Rippling.

Remember when software made business simpler?

Today, the average company runs 100+ apps—each with its own logins, data, and headaches. HR can’t find employee info. IT fights security blind spots. Finance reconciles numbers instead of planning growth.

Our State of Software Sprawl report reveals the true cost of “Software as a Disservice” (SaaD)—and how much time, money, and sanity it’s draining from your teams.

Read the full report and see how your company stacks up→

The future of work is unified. Don’t get SaaD. Get Rippling.

Stop SaaD in its tracks

🤖 Automate Sales, Marketing, and Ops in Minutes with Lindy

Lindy builds AI agents that actually do the work, qualifying leads, sending follow-ups, summarizing chats, drafting reports, and more. No coding, no setup, just describe what you want and Lindy handles it.

Here’s how it works:

1. Tell Lindy the Task: Give a prompt like “Qualify inbound leads” or “Summarize customer chats into HubSpot,” and Lindy builds the full workflow automatically.
   

2. Deploy Your Agent: Your AI agent runs 24/7, researching, drafting, messaging, updating CRMs, and completing repetitive tasks your team normally spends hours on.
   

3. Automate Everything: Connect Lindy to 5,000+ apps to handle booking, reporting, lead nurturing, support triage, and more across your entire stack.

Your next hire isn’t human, it’s an AI agent that never stops working. You can try it here for free

💸 OpenAI’s Leaked Financials Reveal a Bigger Cost Problem

Newly leaked documents reveal major gaps between OpenAI’s publicly reported revenue and its real financial relationship with Microsoft. The data shows lower actual revenue and far higher compute costs than previously believed. 

The Shift:

1. Revenue share reveals lower real revenue - Microsoft received about half a billion from OpenAI in 2024 and nearly $900M through Q3 2025, which implies revenue far below the $4B+ figures reported publicly and exposes a gap between leaked data and OpenAI’s stated annualized revenue.

2. Inference spending has exploded - OpenAI spent nearly $4B on inference in 2024 and over $8.6B in the first nine months of 2025, showing a cost curve far steeper than earlier reports and suggesting the company’s real compute bill massively exceeds industry estimates.

3. Public revenue claims don’t align with the leak - The implied revenue from Microsoft’s 20 percent share contradicts earlier claims of $4.3B in early 2025 and also conflicts with Sam Altman’s statement that OpenAI is making “well more” than $13B annually, making the reporting difficult to reconcile.

OpenAI’s leaked data shows inference costs rising faster than revenue and revealing far tougher economics than public claims. The numbers suggest that if expenses keep outpacing income, the broader AI sector may be facing the same structural strain. This raises deeper questions around sustainability pricing and long-term viability.

_{TOGETHER WITH ROKU}

CTV ads made easy: Black Friday edition

As with any digital ad campaign, the important thing is to reach streaming audiences who will convert. Roku’s self-service Ads Manager stands ready with powerful segmentation and targeting — plus creative upscaling tools that transform existing assets into CTV-ready video ads. Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

Use code GET5K now

🔨AI Tools for the Shift

📝 HumanText.pro – Free AI humanizer that converts AI text into human-like writing and bypasses detectors instantly with no payment required.

🗣 PracticeSales – Practice real sales conversations with AI and get instant feedback to improve faster and sharpen closing skills.

🎥 AI UGC Video Generator – Transform scripts into creator-style UGC videos instantly using AI-powered video generation.

🚀 BlogBuster – Boost visibility everywhere from Google to ChatGPT with AI-optimized content engineered for discoverability.

📱 Sebastian.run – Build professional, polished mobile apps using a platform that blends advanced AI with expert-level development foundations.

🚀Quick Shifts

🗳️ Anthropic detailed how it measures Claude’s political neutrality, using system prompts, reinforcement learning, and a new evaluation tool. Claude Sonnet 4.5 scored 95 percent even-handedness, outperforming GPT-5 and Llama 4.

🤯 Fireflies.ai’s co-founder says the company’s early “AI” was actually just two founders silently joining calls and taking notes by hand, raising major transparency and privacy concerns despite claiming users knew humans were involved.

💰 Google is investing $40B in Texas through 2027 to expand data centers and AI infrastructure, alongside a $30M Energy Impact Fund and new workforce training programs.

🍏 Apple CEO Tim Cook may step down as early as next year, with hardware engineering chief John Ternus emerging as the leading candidate to succeed him, according to reports.

💬 OpenAI has launched a pilot for group chats in Japan, New Zealand, South Korea, and Taiwan, letting up to 20 users collaborate inside shared conversations for the first time.

🤳AI Nugget of the Day

That’s all for today’s edition see you tomorrow as we track down and get you all that matters in the daily AI Shift!

If you loved this edition let us know how much:

Forward it to your pal to give them a daily dose of the shift so they can 👇