Key takeaway: Experienced AI users are shifting toward augmentation over automation. High-tenure users have a 10% higher task success rate and are choosing to keep more human judgment in the loop. The metric most teams track (tasks automated) is the wrong one.
The Anthropic Economic Index, published in March 2026, contains a finding that cuts against the dominant narrative about AI in the workplace. Users with high AI tenure, those who have been using AI tools intensively for the longest time, are shifting their usage patterns toward augmentation and away from automation. They are choosing to keep more human judgment in the loop, not less. And they have a 10% higher task success rate than users who deploy AI in a more fully automated mode.
This is not what the automation narrative predicted. The expected arc was: start with augmentation (human in the loop), gain confidence in the AI, progressively remove the human, achieve full automation. The experienced users are not following that arc. They are settling on a different steady state: one where AI handles specific, well-defined subtasks and humans retain judgment on the synthesis, interpretation, and decision points.
Augmentation vs Automation: The Actual Distinction
Automation in AI means the system completes a task with minimal human input. You give it a goal, it executes, you receive an output. The human role is task initiation and output review, nothing in between.
Augmentation means AI enhances human capability at specific stages of a task while the human remains actively involved in the judgment-intensive steps. The human is not just reviewing the final output. They are engaging with AI assistance at decision points throughout the process, using AI output as input to their own thinking rather than as a replacement for it.
In growth marketing practice, the distinction looks like this. Automated campaign reporting: the agent pulls data, calculates metrics, formats the report, and delivers it. Human reviews the finished report. Augmented campaign analysis: the human asks AI to pull and organise the data, reviews the organised data, forms their own initial interpretation, asks AI to challenge that interpretation, synthesises the exchange into a recommendation. The augmented version takes longer but produces a higher-quality output because the human's contextual knowledge and judgment are embedded throughout, not just applied as a final check.
Why the Wrong Metric Leads to the Wrong Outcome
Most teams measure AI success by tasks removed from the human queue. This metric makes intuitive sense: if AI is automating work that humans used to do, fewer tasks on the human queue is evidence that it is working. The metric is not wrong. It is insufficient.
Tasks removed from queue tells you about throughput. It does not tell you about quality, about whether the important judgment points are being handled well, or about whether the organisation is building capability or eroding it. A team that has automated 80% of their campaign reporting workflow has faster reporting. Whether that reporting is driving better decisions depends on whether the 20% of human involvement remaining is focused on the right things.
The Anthropic finding suggests that experienced users have learned something important: the quality ceiling for fully automated tasks is lower than the quality ceiling for augmented tasks. You can automate a first draft. You can augment a final argument. The first produces faster work. The second produces better work.
What This Looks Like in Growth Marketing Specifically
The augmentation pattern shows up most clearly in three areas of growth marketing work.
Audience analysis
Automated audience analysis: AI segments the database, identifies the highest-value cohorts, generates a targeting recommendation. Human approves. The output is fast and technically correct, but it reflects whatever the AI was trained to optimise for. Augmented: AI pulls the segmentation and surfaces the patterns. Human reviews with knowledge of recent customer conversations, product roadmap, and market context not in the data. The human's contextual knowledge reshapes the interpretation. The targeting recommendation reflects both the data pattern and the strategic context.
Campaign hypothesis development
Automated: AI generates five campaign hypotheses based on past performance data and brief inputs. Human picks one. Augmented: AI generates hypotheses, human challenges each one, AI develops the strongest hypothesis further, human adds domain-specific knowledge about why this matters to this audience now, AI stress-tests the resulting hypothesis. The final hypothesis is significantly sharper than what either produced alone.
Performance interpretation
Automated: AI produces a performance summary. Augmented: AI pulls the performance data and identifies anomalies. Human asks about the anomalies. AI generates possible explanations. Human eliminates explanations that do not fit with contextual knowledge. They converge on an interpretation that leads to a specific, well-reasoned decision about what to change. The interpretation is better, and the resulting decision is more likely to improve future performance.
Building for Augmentation
Designing for augmentation rather than automation requires a different workflow architecture. Instead of building AI flows that minimise human intervention, you design intervention points into the workflow deliberately. At which stages does human judgment most improve the output? Those are the points where you keep the human actively engaged rather than in review mode.
This also changes how you evaluate AI tools. The question is not "how much can this tool do without human involvement?" but "how well does this tool assist human judgment at the specific points where that judgment matters most?" Different answers produce different tool choices and different workflow designs.
Conclusion
Experienced AI users are choosing augmentation over automation, and achieving better outcomes as a result. The 10% higher task success rate is not a small difference. It is evidence that the quality ceiling for augmented work is meaningfully higher than for automated work, at least in the complex, judgment-intensive tasks that define most growth marketing roles. The metric most teams track, tasks automated, measures efficiency. The metric that matters is decision quality, and augmentation produces better decisions because it keeps human judgment embedded in the process rather than applied only at the end.
