Jason On Firms Podcast 586 First Impressions: Claude Cowork for Accounting Firms [Is this the most powerful AI agent for accountants?]
28 snips
Feb 26, 2026 Hands-on testing with Claude Cowork and OpenClaw and what those AI agents mean for accounting workflows. Demos of massive bookkeeping and tax document automation without revealing insights. Security boundaries, agent risks like web access and prompt injection, and when autonomous actions are dangerous. A niche deep dive into high-end custom jewelry firms and practical go-to-market notes.
AI Snips
Chapters
Transcript
Episode notes
Team Rollouts Create Risky Incentive Problems
- Rolling out web-capable agents across a team creates incentive and boundary problems because staff may use autonomous web access for convenience.
- Jason warns the agent's web access feature is tempting and risky, so firms must define clear boundaries before broad deployment.
Start With The Right 1% Of Work For AI
- The transition from humans doing 100% of work to AI taking incremental percentages is messy and requires deciding which small tasks are safe to offload first.
- Jason emphasizes testing the right 1% versus wrong 1% and building boundaries gradually.
Agents Will Become The Primary Software Users
- AI agents could become the primary users of software, turning current apps into slow APIs for agent access.
- Jason notes this layer-shift means software vendors face existential choices: block agents or compete to be agent-friendly.