Hark, founded by Figure AI CEO Brett Adcock, previews Handoff computer-use agent, claims edge over GPT-5.4 and Opus 4.8
The AMW Read
A new computer-use agent entrant from a known robotics founder updates the agent player map with a benchmark claim against frontier labs, but doesn't resolve an open debate or introduce a top-tier lab.
Named counterparties: Figure AI
Hark, founded by Figure AI CEO Brett Adcock, previews Handoff computer-use agent, claims edge over GPT-5.4 and Opus 4.8
Hark previewed Handoff, a computer-use agent the company says outperforms OpenAI's GPT-5.4 and Anthropic's Opus 4.8 on relevant benchmarks. The preview marks Adcock's first public move beyond humanoid robotics into AI agent software, positioning Hark alongside the general-purpose labs it is benchmarking against rather than as a robotics adjacency play.
A credible benchmark claim against two frontier labs' flagship models is a meaningful marker for a company whose public profile to date has centered on hardware ambitions, not model-layer competition. Per the AI Market Watch index, our pipeline logged four items tagged to Hark in the last 90 days against two in the prior 90 window (name-matched, pipeline-ingested sources only) — a pace uptick that lines up with the company shifting from fundraising news toward product news. It also extends a pattern in our prior coverage: Hark's earlier disclosures centered on a large Series A for AI-dedicated hardware, and this preview shows the company building a software agent layer in parallel rather than waiting on the device.
For builders and investors, the signal worth tracking is whether Hark's benchmark claims hold up under independent evaluation and whether Handoff ships as a standalone product or stays tied to future Hark hardware — that distinction determines whether Hark is entering the computer-use agent market as a competitor to incumbent labs or as a differentiator for its own device roadmap.
