OpenAI has always been great at grabbing attention in the news. Their announcements often come with big, bold claims. For example, they announced GPT-2 but said it was too dangerous to release. Or their “12 Days of Christmas” campaign, where they showcased a new product every day for 12 days.
Now, Sam Altman has shared his thoughts on last year, focusing on the dramatic boardroom soap opera around his firing and return. He also made a bold prediction:
“We now know how to build AGI as it’s usually understood. In 2025, we think AI agents will join the workforce and change how companies work.”
AGI (Artificial General Intelligence) means creating an AI that’s as smart and general as a human. Unlike narrow AI, which is built for specific tasks like translating languages, playing chess, or recognizing faces, AGI can handle any intellectual task and adapt across different areas. While I don’t think “AGI is near,” I do believe AI will join the workforce — but maybe not in the way Altman imagines.
Is AGI Near? No, At least Not The AGI We (or Sam) Imagine
The arrival of AGI in 2025 seems very unlikely. Today’s AI, like ChatGPT, works by recognizing patterns and making predictions — not by truly understanding. For example, completing the phrase “Life is like a box of…” with “chocolates” relies on probabilities, not reasoning.
I don’t believe AGI will happen by 2025, and many experts agree. Demis Hassabis, who I worked with at Google, predicts AGI could arrive around 2035. Ray Kurzweil estimates 2032, and Jürgen Schmidhuber, director of IDSIA, suggests closer to 2050. The skeptics are many, and the timeline remains uncertain.
Does It Matter When? AI Is Already Powerful.
Maybe it doesn’t matter exactly when AGI will arrive. Even Sam Altman recently downplayed the “G” in AGI, saying:
“My guess is we will hit AGI sooner than most people think, and it will matter much less.”
I agree with this to some extent. AI already has impressive capabilities. For example, Netflix’s AI knows your movie preferences better than your partner. TikTok’s algorithms have even been joked about for recognizing someone’s sexual orientation before they did. AI excels at pattern recognition—and in many cases, it’s better at it than humans.
Sam Altman Sees That AI “Join The Workforce”
The more important point in Sam’s memo is his belief that AI will “join the workforce.” I completely agree this is going to happen. As I wrote in my AI agent update, for AI to succeed in the workplace, it needs two key things: (1) access to tools and (2) access to data. These are the building blocks for making AI truly effective in enterprise settings. However, even though Sam often links this idea to AGI, it might not be OpenAI leading the charge to provide these AI workforce solutions.
Microsoft’s Pole Position — Access To Users
Who has the workforce tools? Microsoft. Microsoft. Microsoft. They are in pole position. Most people already use Microsoft products — whether they like it or not — and AI is becoming deeply integrated into these tools, with Copilots appearing everywhere.
In 2023 and 2024, many startups launched impressive AI services for office jobs, only to be quickly overshadowed by giants like Microsoft and Google, which have direct access to customers. Take Jasper.ai, for example — a once-celebrated AI tool for drafting text. As I pointed out in this LinkedIn post, similar features are now built directly into Google and Microsoft products, making it increasingly difficult for smaller players to compete.
The Power Of Data Access
AI needs data to be truly effective. If you’re looking for answers about a company’s internal processes or insights from documents, general tools like ChatGPT won’t cut it. What we need are tools that can read and summarize company documents, tailored specifically for enterprise use. As I’ve said before, 2025 will be the year of SEARCH — especially enterprise search. Tools that can answer questions, summarize content, and help users navigate complex information will be game-changers.
Who has access to this kind of data? Microsoft is a big player, but they’re not alone. Salesforce, for instance, holds an enormous trove of valuable data — customer interactions, discussions, process documents, marketing strategies, and more. Does Salesforce want AI agents to help unlock this potential? Absolutely.
It’s no surprise that Salesforce CEO Marc Benioff recently took a jab at Microsoft. He called their AI assistant, Copilot, “disappointing,” saying, “It just doesn’t work, and it doesn’t deliver any level of accuracy.” He even dubbed it “Clippy 2.0” — the funniest insult I’ve heard in a while — before rolling out Salesforce’s own AI solution, Agent Forces.
OpenAI Is “just” The Smartest Tool?
OpenAI doesn’t have the same level of data access or consumer reach as Microsoft, nor does it have Salesforce’s treasure trove of business data. So, what’s their angle? They claim to be the smartest tool on the block — and they probably are, although I personally find Anthropic’s Claude 3.5 currently better than OpenAI’s GPT-4.
OpenAI is betting on their ability to outperform everyone else with superior technology. That’s why Sam Altman confidently claims we’ll see AGI. What’s behind that bold claim? Reasoning — or, as OpenAI calls it, Reasoning.
OpenAI and Reasoning
OpenAI recently launched o1, a model designed to showcase advanced reasoning capabilities through an iterative, self-calling process:
- Iteration and Reflection: The model generates an output, evaluates or critiques it, and refines it in a new round of reasoning.
- Feedback Loop: This creates a feedback loop where the model revisits its outputs, critiques them, and improves them further.
In essence, GPT with o1 doesn’t just provide answers — it plans, critiques the plan, and continuously improves it.
What’s especially noteworthy is the paradigm shift this represents. Instead of simply releasing a bigger model like GPT-5, the next generation of AI models is focused on “thinking longer” during inference. This ability to process iteratively may be what Sam Altman refers to when he says, “We now know how to build AGI.”
Reasoning Is Enough of a Reason?
But does “reasoning” alone get OpenAI into the game? OpenAI still needs access to data and a strong user presence, similar to Salesforce or Microsoft. To address this, OpenAI launched the ChatGPT desktop app for macOS. This app can now read code directly from developer-focused tools like VS Code, Xcode, TextEdit, Terminal, and iTerm2. This means developers no longer need to copy and paste their code into ChatGPT—a common workaround until now. It’s a really useful tool and a smart move to integrate deeper into the developer workflow.
Chatting With Large Language Models Costs Money
Every call to a large language model (LLM) costs money. For heavy ChatGPT users, the $20 subscription might not even cover the cost of their usage. OpenAI recently raised $6.6 billion in a Series E funding round—a much-needed boost to sustain their operations. While Agentforce generates solid revenue from its customers, and Microsoft enjoys a massive financial war chest, OpenAI is still in the early stages of getting businesses and users to pay enough to offset the steep costs of developing cutting-edge AI.
Their $200-per-month premium tier, which includes the expanded version of O1, is a move in this direction. But is it worth the price? Perhaps this is why AGI keeps being part of the conversation—it helps justify the premium positioning. However, the race to create superior models is far from over. Even O1 could soon be outpaced by open-source alternatives, as we’ve seen before with Meta’s Llama.
Speaking of Meta, I’m confident we’ll see their attempts to monetize AI models in 2025. Ultimately, the biggest challenge for these players remains clear: justifying enormous costs without securing a steady and reliable revenue stream.
Sam is right: AI Agents Will Be In The Workforce
In 2025, we’ll see more AI agents entering the workforce, transforming workflows by simplifying, enhancing, and automating tasks across industries. These won’t be all-encompassing AGI models but smaller, specialized models designed for dedicated workflows. AI will scale and improve processes one step at a time, combining traditional AI, context retrieval, and robust user design to tackle challenges like security, hallucinations, and user control.
Success will hinge on delivering value through well-integrated, user-friendly, and ethically designed solutions, as outlined in my framework for building enterprise-ready AI tools. For Sam Altman, the key strategic question won’t be about achieving AGI but about how to price OpenAI’s base models for enterprise customers like Microsoft or Salesforce—especially if OpenAI ends up competing directly with them.
But How Will We Work With Those New AI Colleagues?
Enterprises will emerge as the winners in the race for better models, better data, and better integrations. Their main focus should be on training employees and customers to work effectively with their new AI colleagues. In my eCornell certificate course on AI solutions, I saw firsthand how productivity soared once students learned to communicate with an AI co-pilot. Initially, many struggled to achieve results, but a step-by-step guide on interacting with AI made a significant difference.
Why? Because even with reasoning and planning capabilities, AI isn’t truly “general” yet, no matter how much hype Sam Altman creates. Students had to learn when to rely on the AI and when to apply human judgment. I believe 2025 will be the year companies realize this need and invest heavily in AI education.
