• AI Academy
  • Posts
  • 🤖 OpenAI's Enterprise Report Shows 8× Growth and a Hidden Gap

🤖 OpenAI's Enterprise Report Shows 8× Growth and a Hidden Gap

Plus: OpenAI's Confession System and Google's No-Code Agent Builder

December 11, 2025

Hello AI Enthusiast,

OpenAI dropped an enterprise report, packed with growth numbers and some interesting gaps in the story. They're also experimenting with a "confession" system that teaches AI models to admit when they break rules or take shortcuts. Meanwhile, Google launched Workspace Studio to let anyone build AI agents without coding. Let's break down what these developments actually mean.

Before we dive in: We're hosting a free webinar starting in one hour on AI strategy for executives. We're keeping it tight, 30 minutes covering our TAP™ framework and the AI Tool Stack Map, a simple way to figure out which tools to buy and how they fit together.

The Big Picture 🔊

OpenAI's Enterprise Report Shows 8× Usage Growth

OpenAI's new enterprise AI report reveals ChatGPT now serves 800 million weekly users, with enterprise usage growing 8× over the past year. Workers report saving 40-60 minutes daily, with 75% seeing improved work quality. Structured features like Projects grew 19× year-to-date, showing AI adoption moving beyond simple queries.

Gioele Mottarlini
Gioele MottarliniCOO and Image Addict

Interesting report, though it leans toward marketing rather than research. OpenAI highlights 8× enterprise growth and 30% more messages per worker, but more messages doesn't mean better results, sometimes users keep asking because they're struggling.

The useful insight is the usage gap: top users employ AI 6× more than median workers, while many ignore core features. Real value comes from proper training, not just buying licenses.

OpenAI Introduces "Confessions" for AI Transparency

OpenAI developed a "confessions" technique where AI models produce a separate, honest report admitting when they violated instructions or hallucinated, without being penalized. Tested on GPT-5 Thinking, it caught 95.6% of problematic behaviors. The confession trains solely on honesty while the main answer optimizes for helpfulness. This proof-of-concept aims to surface hidden misbehaviors as models become more capable.

Helin Yontar
Helin YontarCPO and Polyglot

This is addressing a real problem, companies need to trust that AI follows their policies and safety guidelines. The confession system works well for catching when models try to bypass rules or take shortcuts.

The challenge is when a model genuinely believes it's right but isn't, it can't confess to being wrong. Still, it's a step forward for transparency and catching intentional rule-breaking.

Google Launches Workspace Studio for No-Code AI Agents

Google released Workspace Studio, promising anyone can build AI agents without coding. Using Gemini 3, employees can supposedly automate tasks from simple email sorting to complex workflows by describing what they want in natural language. Agents integrate with Gmail, Drive, Chat, and external tools like Asana and Salesforce. Rolling out to business customers over the next few weeks.

Andrea Mattiello
Andrea MattielloCM and Board Lover

About time! The tool looks promising for automating routine tasks and making workflows smoother. The interface is clean and similar to Zapier.

The main limitation is that it's still not flexible enough compared to established automation platforms in the market, and it's heavily focused on Google's own apps.

Bits and Bobs 🗞️

  • The Linux Foundation launched the Agentic AI Foundation to support open source AI agent projects and prevent incompatible, closed systems.

  • Mistral launched Devstral 2, a powerful open-source AI coding model with 123 billion parameters, designed for business use and available via API.

  • Anthropic launched Anthropic Interviewer, an AI-powered tool using Claude, to conduct large-scale automated interviews that reveal professionals’ nuanced views on AI’s impact on their work and productivity.

  • A judge ordered OpenAI to provide millions of anonymized ChatGPT user logs in a copyright dispute, ruling privacy measures are sufficient.

  • The EU is investigating whether Google improperly uses publisher and YouTube content to train its AI models without fair compensation or consent.

  • Slack users can now tag Anthropic's Claude to automatically handle coding tasks using context from conversations and connected code repositories.

  • McDonald's released an AI-generated Christmas ad that sparked debate over AI's role in creativity and concerns about authenticity and job impact.

From Our Founder’s Channels 🤳

Gianluca’s latest LinkedIn post breaks down OpenAI’s enterprise report, highlighting that companies see real ROI only when teams are well-trained, build custom automations, and embed AI deeply into workflows.

LOLgorithms 😂

Let's hope ‘Confessions’ will help us see fewer em dashes around.

That's a wrap on our newsletter! Before you go, here’s a quick recap of our offerings:

  • AI Academy Membership: Get 12 months of access to all our cohort-based programs, live webinars, on-demand courses, and tutorials.

  • AI Agent Bootcamp: Accelerate processes and solve business problems by mastering prompts and building AI Agents, without coding.

  • Practical Introduction to ChatGPT: A free course on using ChatGPT confidently, understanding its workings, and exploring its potential.

  • Customized Corporate Training: Equip your team with the skills they need to unlock the potential of AI in your business.

Catch you next week! 👋