AI agents finish roughly three out of four tasks they are assigned. Users still do not believe them.
Digital Applied published an analysis on June 12 of First Page Sage’s Agentic AI Statistics: 2026 report, a rolling panel study of 8,128 agentic AI users conducted between January 2025 and April 2026. The headline number: a mean 75.3% task completion rate across five major agent platforms, measured on a 487-user performance sub-sample given complex, multi-step tasks.
The more revealing number sits underneath. Of those same users, 54% said they trusted manual search results more than agentic results. Only 34% trusted the agent output more. Among technically sophisticated users, the trust gap in favor of manual search widened to 37 percentage points.
21 Points Separate the Best From the Worst
The study benchmarked five platforms. Cognition Labs’ Devin led at 86% completion. OpenClaw placed second at 81%, followed by OpenAI Agents at 73%, Replit AI Agents at 69%, and Perplexity Computer at 65%. That 21-point spread between top and bottom means agent selection alone adds or removes roughly a fifth of completed tasks.
The platform rankings invert when measured by usage. According to the study, OpenAI Agents leads with 2.7 million monthly active users in Q1 2026, followed by OpenClaw at 2.3 million. Devin, the completion leader, has 329,000. Popularity and reliability are tracking on different axes.
Citation Depth Correlates With the Trust Deficit
The analysis highlights a pattern that connects completion rates to the trust gap: agents that cite more sources per task carry lower trust risk with users, regardless of how fast they finish.
OpenClaw cited a median of 7 sources per task, ranging from 3 to 15. Devin and OpenAI Agents each cited a median of 2 sources, with ranges of 1 to 4. Replit AI Agents fell in between at 5. According to Digital Applied, “users who can evaluate a citation trail notice when there isn’t one.”
Only 18% of successfully completed tasks required any user follow-up, and agents delivered an average 66.8% time savings across task types. Trip planning saw the largest gain: 9.2 minutes with an agent versus 38.5 minutes manually, according to First Page Sage.
Methodology Caveats
Digital Applied flags several limitations. The 75.3% completion figure comes from a 487-user sub-sample, not the full 8,128 panel. The study is vendor-commissioned by First Page Sage and has not been independently peer-reviewed. The analysis treats the numbers as “one careful read of the market rather than settled industry facts.”
The Metric That Matters
The study reframes a question that agent builders have been optimizing around for the past year. Completion rate measures whether an agent finished a task. It says nothing about whether the user believed the result.
For teams building on agent platforms, the practical implication is specific: an evidence trail of checkable sources converts task completion into user confidence. Without it, a fast agent that finishes 86% of tasks can still lose the trust of the users most valuable to retain. The agents scoring highest on citation depth in this study are the ones with the lowest trust risk, even when they finish fewer tasks overall.