Contrary to the vision of free-wheeling autonomous agents, most business automation relies on strict Standard Operating Procedures (SOPs). Products like OpenAI's Agent Builder succeed by providing deterministic, node-based workflows that enforce business logic, which is more valuable than pure autonomy.

To successfully automate complex workflows with AI, product teams must go beyond traditional discovery. A "forward-deployed PM" works on-site with customers, directly observing workflows and tweaking AI parameters like context windows and embeddings in real-time to achieve flawless automation.

To build a useful multi-agent AI system, model the agents after your existing human team. Create specialized agents for distinct roles like 'approvals,' 'document drafting,' or 'administration' to replicate and automate a proven workflow, rather than designing a monolithic, abstract AI.

Counterintuitively, the path to full automation isn't just analyzing conversation transcripts. Cresta's CEO found that you must first observe and instrument what human agents are doing on their desktops—navigating legacy systems and UIs—to truly understand and automate the complete workflow.

The effectiveness of enterprise AI agents is limited not by data access, but by the absence of context for *why* decisions were made. 'Context graphs' aim to solve this by capturing 'decision traces'—exceptions, precedents, and overrides that currently live in Slack threads and employee's heads, creating a true source of truth for automation.

Training AI agents to execute multi-step business workflows demands a new data paradigm. Companies create reinforcement learning (RL) environments—mini world models of business processes—where agents learn by attempting tasks, a more advanced method than simple prompt-completion training (SFT/RLHF).

Instead of pre-engineering tool integrations, Block lets its AI agent Goose learn by doing. Successful user-driven workflows can be saved as shareable "recipes," allowing emergent capabilities to be captured and scaled. They found the agent is more capable this way than if they tried to make tools "Goose-friendly."

The most significant gains from AI will not come from automating existing human tasks. Instead, value is unlocked by allowing AI agents to develop entirely new, non-human processes to achieve goals. This requires a shift from process mapping to goal-oriented process invention.

To build coordinated AI agent systems, firms must first extract siloed operational knowledge. This involves not just digitizing documents but systematically observing employee actions like browser clicks and phone calls to capture unwritten processes, turning this tacit knowledge into usable context for AI.