When Jesal Gadhia cofounded a software company a year ago, he expected that the AI agents it was creating would save its customers a lot of work.
He didn’t predict the same tools would save his own team at the startup Cora so much time.
Agents wrote all of the code the company uses — something that wouldn’t have been possible before last year, he told Business Insider.
The company’s six-person team produced what Gadhia calls “unprecedented” amounts of code in its first 12 months. Five years ago, he said, reaching the same level of productivity would have required 20 to 30 engineers.
“It’s been unlike any other time that I can remember,” Gadhia said of the impact of agents.
Across tech, AI agents powered by large language models are absorbing tasks that experienced engineers once handled. Software engineering is becoming a human-AI partnership — what Anthropic chief Dario Amodei has called the industry’s “centaur phase.” And, as some tech insiders increasingly warn, what begins in software rarely stays there, with potential implications for other white-collar fields.
An overnight assistant
At Canva, the graphic design software company, engineering teams draft detailed instructions for AI agents to execute in the background — sometimes overnight. By morning, the work is ready, Brendan Humphreys, Canva’s chief technology officer, said.
“Often, those results are really impressive,” he told Business Insider.
Engineers still apply a “human touch” to reach the company’s quality bar. Even so, agents are delivering “hours and hours and hours of work done completely autonomously,” Humphreys said.
That’s changing what it means to be a coder.
Humphreys said that his senior engineers now often describe their jobs as “largely review” — checking AI output, steering one or more agents to follow a plan, and taking responsibility for the final product.
Teams still spend time defining problems.
“The hardest part of engineering is to translate often vague, confusing, conflicting requirements into something that is production-ready,” he said.
AI can help, but doing it well requires “precision of articulation” in what’s required, Humphreys said. It also demands “mastery of the domain” so engineers can quickly verify that what AI produces is correct — and prevent unnecessary complexity from creeping into Canva’s roughly 70 million lines of code, he said.
“These tools can have you in a jungle before you know it,” Humphreys said of agents.
Higher-level work
At Cora, Gadhia compares AI to a typewriter: It generates the code, freeing engineers to focus on “higher-level strategic architecture,” meeting customers, and brainstorming features.
Cora builds agents that help software companies manage customer relationships. The agents take on tasks like gathering customer requirements, drafting presentations, and following up with clients, he said.
The AI will “run around, do all this work, and you can supervise them,” said Gadhia, who is also the San Francisco company’s chief technology officer.
Agents are also lowering technical barriers. Gadhia said Cora’s CEO, who doesn’t have a technical background, recently asked an agent to change the font on the company’s website during a redesign. Minutes later, after an engineer reviewed the agent’s work, the site was updated.
As agents handle more tasks — something that appeals to some, but rankles others — debate inside tech over AI has intensified.
Microsoft’s AI chief, Mustafa Suleyman, warned in a recent interview that the technology will be able to handle “most, if not all, professional tasks” within 12 to 18 months. AI observers are divided over how disruptive the technology will ultimately be, with some forecasting a massive fallout for desk workers and others saying such fears are overblown.
Some investors are growing cautious. Stocks in industries potentially exposed to having AI wash over profit centers — from finance to software to legal services — have taken hits.
‘AI will sing for you’
Even as agents become more powerful, they’re unlikely to replace entire roles in various industries overnight. For one reason, technical challenges like hallucinations continue.
At the same time, many companies are still figuring out where AI fits into workflows, how workers should validate its output, and how organizations need to adapt, said Muqsit Ashraf, group chief executive of strategy at Accenture. There is often still a role for humans, he told Business Insider.
“Technology for the sake of technology doesn’t help,” Ashraf said.
Fewer than one in 10 organizations has redesigned jobs to support AI adoption, Accenture found in surveys of leaders and workers in 20 countries during the final months of 2025. That’s despite the share of organizations using agents across multiple functions rising to 31% from 27% in a mid-2025 snapshot.
Alex Salazar, cofounder and CEO of AI infrastructure startup Arcade, said that to make the most of agents, workers should treat them like junior employees. That means telling the AI what to do, providing the criteria for success, and, if possible, providing examples.
Do those three things and “AI will sing for you,” Salazar said.
He describes the workplace shift around AI bluntly. As it grows more capable, he said, workers such as software engineers will need to continually redefine their roles.
AI is “improving at an exponential rate,” Salazar said. “And you, as a human, are not.”
Do you have a story to share about how AI is changing your job? Contact this reporter at tparadis@businessinsider.com
