It's one thing for a market leader to sell a vision of a transformative new technology; it's another thing entirely when they turn the lens inward to see how that technology is actually changing their own workplace.
That’s exactly what AI contender Anthropic did, surveying its own engineers and researchers on how they use their powerful AI system, Claude. The results pull back the curtain on the subtle, often contradictory, ways AI is impacting the daily duties of tech workers—and offer critical insights for any business considering an AI rollout.
The Double-Edged Sword of Productivity
Anthropic's analysis shows that AI is driving "significant transformations," primarily by helping engineers "get a lot more done." This aligns perfectly with the bold promise of AI: taking on mundane tasks to free up workers for high-level productivity. Claude is accelerating learning, speeding up iteration, and even allowing staff to tackle previously neglected work.
However, a closer look reveals a paradox:
| The Gain (Hope) | The Trade-Off (Concern) |
| More Done: AI takes on low-level tasks, increasing overall output. | Loss of Competence: Workers worry about losing deep technical skills. |
| Accelerated Learning: Engineers can tackle wholly new tasks outside their core expertise. | Cognitive Overhead: Debugging and cleaning up AI-generated code requires more mental effort than writing it from scratch. |
| Persistence: AI helps workers "grind away" and successfully complete tasks they might have given up on. | Decreased Collaboration: More AI collaboration leads to less interaction with human colleagues. |
Perhaps the most fascinating finding was the split reaction on time savings. Some engineers reported that Claude saved them time, but others said they "spend significantly more time on tasks that are Claude-assisted."
Why the difference? Because managing the output of an AI requires a new kind of work:
Debugging and Cleanup: AI-generated code often needs significant correction.
Supervision and Oversight: Engineers must "shoulder more cognitive overhead for understanding Claude’s code" before they can deploy it.
A New Way to Measure Success
This internal research at Anthropic validates what other studies have suggested: AI is not a simple productivity booster. It fundamentally changes the nature of the work itself.
For your office, the big takeaway is clear: don't expect AI to be a panacea, and don't rely solely on time-based metrics.
If you see your employees spending longer on certain tasks after an AI implementation, it’s not necessarily a sign of failure. It could mean they are:
Self-Educating: They are using the AI to learn new skills and tackle work previously outside their range.
Thinking Expansively: They are shifting their focus to higher-level strategic problems that the AI has freed them to address.
Refining Outputs: They are doing the crucial work of ensuring the AI’s output is accurate and ready for deployment.
Instead of just checking the clock, take a page from Anthropic's book: interview your staff. Ask them about their actual experience. You may find that the initial investment in time is paying off in long-term skill building and strategic capacity—a boon far more valuable than marginal time savings.