Work Units
The operating map of AI-assisted work: what humans and agents are doing, where it repeats, what leaks value, and what should improve next.
distinct units observed this period · 74% mapped to an owner or roadmap task
of all Work Units recur · $14.6k of that runs without memory
lost to retries & duplicate effort · concentrated in 6 units
drafted from recurring patterns · 3 ready for review now
leaks in repeated work this period. Recommended next move: capture the escalation answer asset and break the retry loop.
Work Unit operating map
Where the work changes hands
Each Work Unit is a relay between human judgment and agent labor. The handoff points are where quality is won and where leaks hide.
Reads the escalation, frames the ask
judgmentSummarizes ticket history, drafts a response
4,100-token context pasteEdits tone, checks policy edge cases
Searches prior cases, attaches citations
Final call and send
the decision stays humanWhat changed
Meaningful movement in Work Units, memory, leakage, and agent behavior — the same live feed as Home, scoped to work.
- 1m ago
Corey RiceProduct & AIWorkClaude CodeWiring the Gemini Web capture parser
138 classified exchanges today on the capture stack. Matched to the Gemini Web capture Work Unit with high confidence.
- 2m ago
Yonatan HaimovitzProduct & AIWorkCodex / Wham“is this the right fix for staging? is this the right way to connect to postgres?”
6-turn session on the incident Postgres fix. Tool schemas cost 4.4k tokens on every call — flagged for the harness manifest.
- 4m ago
Omer GoldbergLeadership · CEOWorkClaude CodeApplying the SEN-12 tooltip persistence patch
Validating expanded-mode UI in the Cortex sensor dashboard. Retrieval pulled a deprecated handbook section — answer was human-corrected.
- 8m ago
Teo DeleanuProduct & AIMemoryCursor IDEReused the provider-capture runbook instead of re-deriving it
His Knowledge Object answered before a model ran — now 38 reuses this month, the most-reused object in the org.
- 12m agoCortexPattern detectionCortex noticed
Pattern detected — refund-exception answers drafted from scratch
Support has re-derived the same refund-exception response 14 times this week. This is a Knowledge Object candidate.
- 18m ago
Paolo FacchinettiProduct & AI · DevOpsWorkClaude Code“Deploy Gemma on EKS with vLLM — cleanest path for the inference service?”
Answered straight from the deploy runbook Knowledge Object. Context load 5× below the org median — the pattern to converge on.
Top recurring Work Units
by occurrences · 30DRepeated work without memory is the bill that keeps coming back. See the patterns →
Who and what is powering the work
Work Units are created by humans, agents, tools, and projects working together. Contribution evidence, not productivity ranking.
This is not a leaderboard. It is a map of where AI-assisted work is happening and which people, agents, and systems are involved.
Work by team
Repeated work without memory is the bill that keeps coming back. Every drafted candidate waits in the Memory review queue. Browse all Work Units



