Pragmatiq AI News — AI Trends and Business Insights in Estonia

How AI is Transforming Work: Real Numbers from Anthropic Research

2025-12-11 13:03
Anthropic conducted a large-scale internal study to understand how artificial intelligence is changing the way we work. In August 2025, they surveyed 132 engineers and researchers, conducted 53 in-depth interviews, and analyzed 200,000 real Claude Code sessions.

Key Numbers

  • +50% productivity — average employee productivity gain.
  • 59% of work time — share of tasks using AI (28% a year ago).
  • 27% of work — tasks that wouldn't be done at all without AI.
  • 67% increase in merged pull requests per engineer per day.
14% of employees more than doubled their productivity — these "power users" mastered effective AI collaboration strategies.

What Tasks Are Delegated to AI Most Often

Figure 1: Proportion of employees using Claude daily for various programming tasks

Most popular tasks:
  • Debugging — 55% use daily.
  • Code understanding — 42% daily.
  • Implementing new features — 37% daily.

How Productivity Actually Grows

The study revealed an interesting pattern: time spent on tasks decreases slightly, but work output volume increases substantially.

Figure 2: Claude's impact on time and work output. Most tasks require less time but produce much greater output volume

What this means in practice:
  • Employees tackle tasks they previously postponed.
  • Do more thorough testing.
  • Improve code quality where time was lacking.
  • Explore new approaches and experiment more.
"People tend to think about super capable models as a single instance, like getting a faster car. But having a million horses… allows you to test a bunch of different ideas… It's exciting and more creative when you have that extra breadth to explore." — Anthropic researcher

Trust Progression: From Simple to Complex

One of the most interesting patterns is how employees gradually expand their AI use. Many described delegating increasingly complex tasks over time.
One engineer compared it to getting used to Google Maps:
"In the beginning I would use Google Maps only for routes I didn't know... This is like me using Claude to write SQL that I didn't know, but not asking it to write Python that I did. Then I started using Google Maps on routes that I mostly knew, but maybe I didn't know the last mile... Today I use Google Maps all the time, even for my daily commute. If it says to take a different way I do, and just trust that it considered all options... I use Claude Code in a similar way today."
A security engineer emphasized the importance of experience when evaluating AI suggestions: Claude proposed a solution that was "really smart in the dangerous way, the kind of thing a very talented junior engineer might propose." Only an experienced user could recognize the potential problem.

Effective Delegation Criteria

✓ Successful AI users delegate tasks that are:
  • Easy to verify — where validation effort is small compared to creation.
  • Outside core expertise — low complexity but require unfamiliar technologies.
  • Well-defined — subtasks that can be isolated from the rest of the project.
  • Quality not critical — temporary debugging or research code.
  • Boring or repetitive — 44% of AI-assisted work consists of tasks employees wouldn't want to do themselves.
✗ NOT delegated:
  • High-level strategic thinking.
  • Design decisions requiring organizational context.
  • Tasks where experience and "taste" matter.

The Full-Stack Revolution: Expanding Competencies

Employees have significantly expanded their capabilities:
"I can very capably work on front-end, or transactional databases... where previously I would've been scared to touch stuff"
A backend engineer created a complex UI with Claude's help:
"It did a way better job than I ever would've. I would not have been able to do it, definitely not on time...The designers were like 'wait, you did this?' I said "No, Claude did this - I just prompted it.'"

AI Becomes More Autonomous

Analysis of 200,000 real sessions shows dramatic changes over six months:

Figure 3: Over 6 months, task complexity increased, actions without human intervention grew 116%, human interactions decreased 33%

  • Task complexity: 3.2 → 3.8 (on a 1-5 scale).
  • Consecutive actions without intervention: ~10 → ~21 (+116%).
  • Human interactions per task: 6.2 → 4.1 (-33%).

Figure 4: Significant growth in usage for implementing new features (14% → 37%) and design (1% → 10%)

How Different Teams Use AI

Figure 5: Each team uses AI differently, expanding their core competencies

  • Security team: 49% — analyzing unfamiliar code.
  • Researchers: 7-8% — frontend for data visualization.
  • Non-technical employees: 52% — debugging, 13% — data analysis.
8.6% of all tasks are "papercut fixes": small quality-of-life improvements (refactoring, creating tools) that were previously postponed but now accumulate into significant impact.

The Meaning of Work is Changing

Engineers sharply diverge on losing hands-on coding. Some feel genuine loss:
"It's the end of an era for me - I've been programming for 25 years, and feeling competent in that skill set is a core part of my professional satisfaction."
Others accepted the trade-off:
"There are certainly some parts of writing code that I miss - getting into a zen flow state when refactoring code, but overall I'm so much more productive now that I'll gladly give that up."
Still others focus on outcomes:
"I expected that by this point I would feel scared or bored… however I don't really feel either of those things. Instead I feel quite excited that I can do significantly more. I thought that I really enjoyed writing code, and instead I actually just enjoy what I get out of writing code."
One engineer challenged the premise itself:
"The 'getting rusty' framing relies on an assumption that coding will someday go back to the way it was pre-Claude 3.5. And I don't think it will."

Main Employee Concerns

1. Skill Atrophy

"When producing output is so easy and fast, it gets harder and harder to actually take the time to learn something."
Manual problem-solving provides "incidental learning" — when debugging, you read documentation and code, building a mental model of the system. With AI, this process diminishes.

2. The Supervision Paradox

Effective AI use requires checking its work. But checking requires skills that atrophy from AI overuse.
"I'm primarily using AI in cases where I know what the answer should be or should look like. I developed that ability by doing SWE 'the hard way'... But if I were earlier in my career, I would think it would take a lot of deliberate effort to continue growing my own abilities rather than blindly accepting the model output."

3. Changing Social Dynamics

Half of respondents reported less colleague interaction. Claude became the first stop for questions: "80-90% of questions go to Claude."
Mentorship suffers too:
"It's been sad that more junior people don't come to me with questions as often, though they definitely get their questions answered more effectively and learn faster."

4. Career Uncertainty

Many see their role shifting from writing code to managing AI agents. Some expressed conflict between short-term optimism and long-term uncertainty:
"I feel optimistic in the short term but in the long term I think AI will end up doing everything and make me and many others irrelevant."

How to Adapt

For Employees:
  • Develop strategic thinking
  • Practice deliberately without AI
  • Use AI for career development
  • Be adaptable
For Leaders:
  • Invest in training for effective AI work — the difference between average and power users is huge
  • Create conditions for practice without AI — protection against critical skill atrophy
  • Rethink mentorship processes — traditional approaches are changing
  • Prepare for role evolution — "AI agent manager" role is becoming reality
  • Pay attention to social dynamics — reduced colleague interaction requires conscious approach

What This Means for Other Industries

While the study was conducted at a tech company with early access to cutting-edge tools, the patterns apply more broadly. Key insights are relevant for any business:
Productivity grows through volume, not speed — people do more, not faster.
This means AI's value isn't in replacing people, but in expanding their capabilities.
Delegation requires strategy — not everything should go to AI.
Successful users develop intuition about which tasks to delegate and which to keep.
Supervision skills become critical — the ability to check and guide AI work requires deep domain understanding.
This is the paradox: AI can weaken the very skills needed to use it effectively.
Social aspects matter — changes in team dynamics, mentorship, and career trajectories require attention no less than technical implementation aspects.

Business Takeaways

  1. Real productivity growth is possible — but requires strategy. The difference between average user (+50%) and power user (+100%+) is in delegation skills.
  2. New work, not replacement — 27% of work wouldn't be done without AI. This is additional value.
  3. Supervision skills are critical — effectiveness requires checking ability, not blind trust.
  4. Early adoption provides advantage — over 6 months of the study, models became significantly more capable.
  5. Prepare for uncertainty — even AI creators say: "Nobody knows what's going to happen… the important thing is to just be really adaptable."

Want to Understand How AI Can Transform Your Company?

The numbers from Anthropic's research are impressive, but every business is unique. Which processes in your company can be automated? Where will AI provide maximum ROI? What risks should be considered?
Book a free consultation — we'll help assess AI's potential for your business and create an implementation plan tailored to your industry.

[Book a free consultation]

Based on Anthropic's research "How AI Is Transforming Work at Anthropic" (December 2025). All charts and images are used from the original Anthropic material.