SerpApi.Org
Sol Rashidi helped launch IBM’s Watson reveals why half her agents failed—and what most companies still get wrong about scaling AI.
If anyone on earth should be bullish on AI agents, it’s Sol Rashidi. She’s spent 15 years deploying AI inside Fortune 100 companies, becoming one of the world’s first chief AI officer in 2016 and helping IBM launch Watson in 2011.
That’s why I was surprised when I learned Rashidi has fired half of her agentic staff: “The ones I fired were unreliable,” she told me. They were unpredictable, and it was actually taking me more time managing them and course correcting than it would for me to actually train an early career adult or a college graduate.”
Read that again. The person who helped bring Watson to market is firing her own AI agents.
“It’s easy to build an agent if you’re a team of one or perhaps seven,” she says. But at the enterprise level, she says, adding that only one in three agents performs well consistently.
That’s not the story you see on LinkedIn. And that gap between the hype and the reality is exactly what Rashidi wants everyone to understand.
Stop running with scissors.
“We’ve created a market where everyone’s being told it’s okay to run with scissors, but the reality of AI is, it’s hard work when you want it to scale predictably and safely,” Rashidi told me.
She has watched this movie from the beginning. When Watson beat Ken Jennings on Jeopardy, it was one of the first times the world saw human against machine. But she reminded me that the original intent was never replacement.
AI was introduced to the market with the intent to amplify people, to automate the monotonous work so humans could do what humans do best. Somewhere along the way, the narrative became: deploy AI, which leads to less headcount and, therefore, a healthier P&L.
The data, she says, tells a messier story. At the macro level, major consulting studies show that most companies are frustrated they can’t extract real value from AI. At the individual level, AI helps with familiar tasks until employees hit what she calls the escape-velocity zone, where they spend more time babysitting the AI’s mistakes than doing the work itself.
“If you’re using Copilot, Gemini, OpenAI to summarize emails and take meeting notes, that’s not where you’re going to extract the real ROI of AI. While it can facilitate that stuff day to day, it’s not why this stuff was created,” she says.
Where AI really helps.
Rashidi’s framework for who should go all in stuck with me. She breaks companies into startups, scale-ups, and grownups.
If you’re a team of five with pre-seed funding, AI is a genuine advantage because the stakes are low and you’re building tech-native. But once you’re a scale-up doing $200 million in revenue, the reality is, you’re not going to grow by managing 500 agents. And Fortune 500s — the grownups — are built on decades of customer trust that one rogue agent can torch in an afternoon.
One-size-fits-all is a myth. Your AI strategy should match your stage, not your feed.
A CEO friend of hers, someone running a company doing $100 million to $150 million a year, has been uploading his full health records to a chatbot because he likes the answers better than his doctor’s. No masking. Name and birthday included.
As she put it, the whole market is operating on one philosophy: “I’d rather get speeding tickets than parking tickets.”
This isn’t just podcast talk for Rashidi. Two weeks before our conversation, she spoke at U.S. AI Congress pushing for what she calls bumpers, not brakes. Her favorite idea is borrowed from the grocery store: a nutrition label for AI. Just as the FDA forced food companies to disclose ingredients, AI companies would disclose what data trained their models, what was excluded, and what biases were observed. Then people can choose.
And this month, Harvard Kennedy School published Rashidi’s working paper on the future of work, warning that AI is automating the exact entry-level jobs where young professionals have always built judgment and expertise. She calls the compounding risk intellectual atrophy. The employment data for workers ages 22 to 25 is already flashing yellow.
Which brings me to her secret sauce, and it’s not a tool or a prompt.
Rashidi isn’t anti-AI. She built her career on it. Her edge is that she treats AI like the double-edged sword it is: magnificent when done right, dangerous when adopted out of FOMO.
The entrepreneurs who win won’t be the ones who adopt AI fastest. They’ll be the ones who adopt it deliberately, with their eyes open, at the right stage, for the right reasons.
Everyone else is just running with scissors.
This post originally appeared at inc.com.
“Click here to subscribe to the Inc. newsletter: inc.com/newsletters"
