Unsupervised — Intelligence Is No Longer the Bottleneck
Intelligence is no longer the bottleneck.
With appropriate supervision, it's clear that current and future models can autonomously create incredible things. We build tools and systems for higher abstractions of human supervision in agentic engineering and agentic data analysis.
Results
Already at work for real customers.
$1B+ in actionable insights found by customers since 2021
$100M+ in estimated value from insights AT&T put into action
$58M found for a Fortune 500 healthcare payer in six months
65.7% fewer hallucinations than peer tools in natural-language data queries
DA-Bench
36% fewer zero-score failures when the same agent runs DeepWork workflows
AgentIF-OneDay
2.4× tighter run-to-run consistency for the same agent on DeepWork workflows
AgentIF-OneDay
One run, from its own logs: kicked off at 11:24 PM, searched 194 million rows overnight, got its draft bounced by a quality gate, and had a human-approved report by 6:05 AM.
run · hhs-medicaid
06:24:20 job accepted analyze medicaid FFS spending · scope: 8 questions
06:43:48 analysis done 4 parallel agents · external sources verified
06:44:44 data exported 194,002,693 rows (2018–2023) → parquet
07:30:24 patterns found 9,588 candidates · 2,174s · 30 workers
12:37:22 patterns ranked 300 kept by KS score · 0 errors, 0 warnings
12:58:38 quality gate report drafted · gate passed on attempt 2
13:04:57 approved user approved the report · workflow complete
Actual event log · timestamps in UTC · Unsupervised.com — March 2026
Products
Trust what your agents do.
Agent Control Systems
DeepWork Give agents workflows they can follow without constant supervision.
Panopticon Watch unattended runs by exception.
Outfitter Dotfiles for your coding agent: make, share, and switch profiles.
2119 Enforce that every requirement has an honest test.
Agentic Data Analysis
Finder Give agents evidence before they act.
Finder for Teams
See it in action.
This is where we learned to control agents in production. It runs against warehouse data: KPI-linked patterns, ranked by estimated financial impact, with evidence an analyst reviews before anyone acts.
Explore Finder for Teams Request a demo →
Named customer · verified
“Unsupervised’s AI Data Analysts have delivered strong ROI—improving our key metrics while empowering our team with faster, smarter access to data insights.”
Mark Austin · VP Data Science, AT&T
Writing
Follow our work.
Research, field notes, and product updates for engineers and analysts working with agents.
Cost readout Structure makes frontier models more expensive and open-weights models cheaperRead → Early readout Workflows written by a frontier model made an open model twice as fastRead → Example HHS Medicaid analysis: a full run, publishedRead → Case study AT&T put $100M+ of insights into actionRead → Research SemLang: a semantic language that gives agents a governed view of your dataRead →