From Chat to Workflow Agents

The real shift is from asking AI questions to having AI complete packaged tasks inside the tool people already use. In enterprise work, the expensive part is not finding an answer, it is turning many documents into a memo, diligence grid, contract review, or pitch deck with the right format, permissions, and audit trail. That makes chat a useful entry point, but a weak final interface for repeated, high stakes workflows.

• Hebbia’s product direction makes this concrete. Matrix is a spreadsheet style agent layer that takes dropped files, CRM data, SharePoint docs, and research sources, then runs multi step reasoning to produce outputs like meeting notes, diligence workbooks, and memos. Chat still exists in Lite, but the core system routes work through the workflow engine.
  1 sacra 2 sacra
• The best comparison is Glean versus Hebbia. Glean won early by making enterprise knowledge searchable and chat friendly across many apps, and reached $100M ARR in November 2024. Hebbia charges far more, starting at $3,000 to $3,500 per seat per year for Lite and $10,000 per seat per year for full access, because it is selling completion of finance and legal work, not broad companywide chat.
  1 sacra 3 sacra 4 sacra
• This is not unique to one company. Harvey also shifted from a specialized legal model toward agentic workflows, and both companies pair software with ex practitioners who configure templates and drive change management. The pattern is that enterprise AI value moves toward opinionated workflows, services, and governance, where generic chat is easiest for a platform model provider to copy.
  1 sacra 4 sacra 5 sacra

Over the next few years, enterprise AI products are likely to look less like a chatbot and more like a set of prebuilt work cells, automations, and review flows for each team. The winners will be the companies that can encode a department’s exact way of working, prove accuracy with citations and audit logs, and turn occasional prompting into daily system of record behavior.