👀 Five contrarian ideas about genAI in the workplace

Everyone knows of AI, but do they *know* AI?

Hi all,

By now, AI is on everyone’s radar. ChatGPT alone sees over 300 million weekly users—roughly 7% of all mobile phone owners worldwide. Nearly a third of knowledge workers harness AI, but there’s still a big gap between being aware of AI and genuinely understanding it.

Here are five contrarian ideas that may just change your perspective—and your strategy—on AI at work. But before we start, here’s a quick reminder…

---

Special member offer: As we start the new year, it’s the best opportunity to get Azeem into your organisation to bring the insights of Exponential View to your team and clients. These spots book up quickly but we prioritise members of Exponential View. Get in touch to discuss the opportunities for speaking, board meetings and customer conferences.

---

1. You're likely only scratching the surface

What the Zeitgeist says: We’re already using AI. We use it to draft emails, summarise reports and retrieve facts. That’s real AI productivity!

What’s really going on: Those tasks barely skim the top – and there’s a growing differentiation between those who have ‘buy-in’ and those who have ‘belief’ in AI. ‘Buy in’ means performative apps, box ticking, the customer service bot is the perfect example. ‘Belief’ means retooling entire processes around AI.

Early adopters are building entirely new ways of working. Synthetic agents are already in use to chain tasks autonomously and pursue long-term objectives. McKinsey forecasts that agents may automate up to 30% of tasks by 2030, but the takeover may happen sooner than that. Jensen Huang spoke about how Nvidia uses agents and that’s what belief looks like:

Jensen: Our cybersecurity system today can’t run without our own agents. We have AI agents helping design chips—Hopper wouldn’t be possible, Blackwell wouldn’t be possible and don’t even think about Rubin. We have AI chip designers, AI software engineers and AI verification engineers, and we build them all internally. We have the ability, and we’d rather use the opportunity to explore the technology ourselves. I’m hoping that Nvidia someday will be a 50,000 employee company with a 100 million AI assistants […] AIs will recruit other AIs to solve problems […] and so we’ll just be one large employee base, some of them digital and some biological.

Dig deeper: 2025 will be the year of agents

---

2. AI isn’t the great equaliser

What the Zeitgeist says: AI boosts everyone’s performance equally. If you’re weaker at a skill, AI helps you more, narrowing the gap with top performers.

What’s really going on: Early studies of AI assistants (not fully autonomous agents) supported the assumption that less-skilled workers caught up with their higher-performing peers. But the latest research that takes into account the rise of agents is changing the expectations. A paper by Ide and Talamas suggests that skilled professionals could quickly automate menial tasks and double down on strategic work, while novices struggle to deploy these systems effectively.

Separately, even with Assistants, Otis et al. found that high-performing entrepreneurs saw 15% profit gains with AI, while weaker peers lost 8%—a remarkable divergence. The reason? The best can better select and implement suggestions. If these trends hold, those at the top will stretch their lead, whether through adroit delegation to agents or better judgement when sifting AI-generated advice.

---

---

3. Bans don’t work

What the Zeitgeist says: AI is risky so a simple ban in the workplace will keep it out.

What’s really going on: Bans don’t work. A hefty share of employees—especially younger ones—admit to using AI tools behind the scenes. A recent survey found a third of workers share confidential info with AI without permission.

Your staff won’t wait for the official approval. It’s in your best interest to encourage those who want to experiment. There’s some early evidence to back this up: Bain & Company reckons integrating genAI could lift earnings by as much as 20% in just 18-36 months.

---

4. The cost-performance curve is shifting fast

What the Zeitgeist says: AI is impressive, but is this it? It still feels costly and slow compared with real human ingenuity.

What’s really going on: That gap is closing fast. Llama 3.1 can handle around 1,700 tokens per second—hundreds of times faster than a human reader. Testing shows that agents can complete certain tasks at just 3% of a baseline human cost. The contrarian question is not “Wow, ChatGPT is cool—should we adopt it?” but rather “Is this really all AI can do, or is the cost-performance curve about to unlock even more profound disruptions?”

OpenAI’s newest o3 model has shown unprecedented adaptability to complete new kinds of tasks in test environments. AI is only going to get faster, cheaper and more capable.

Dig deeper: Beyond human – OpenAI’s o3 wake-up call

---

5. Hallucinations don’t matter any longer (as much as you think)

What the Zeitgeist says: AI can’t be trusted. It hallucinates and spews inaccurate information.

What’s really going on: It is true that AI models sometimes invent facts. DeepMind’s FACTS Grounding benchmark shows that even the best model fails 17% of the time. That is obviously unacceptable for high-stakes domains like medicine or finance. But it is possible to slash certain types of errors. LinkedIn’s engineering team cut down their model’s API formatting errors from 10% to 0.01% with targeted refinements. You can’t look at the models in isolation. They will be a part of systems with guardrails that can reduce their errors.

---

Most workplaces are only skimming the surface of what AI can do, and the pace of change is accelerating. Bans won’t halt its spread, early adopters are using it to pull ahead and hallucination concerns are increasingly surmountable. So carve out the time to truly understand AI —it’s evolving beyond just a tool, edging ever closer to being your future coworker.