Owning the Assistant or Workflow

The core split is between owning the assistant and owning the workflow layer. Adept is trying to make software already used at work more operable by AI, so the product value comes from clicking through enterprise tools reliably. Inflection took the opposite path, building Pi as a single companion with its own model and voice, where the moat is user affinity and model quality rather than deep integration across many apps or agents.

• Adept is the most direct example of adapting AI to existing software. Its product is built around turning natural language into actions inside the tools a company already uses, and its workflow demos emphasize UI understanding, demonstrations, and cross software execution rather than replacing the whole computing environment.
  1 sacra 2 adept 3 adept
• Inflection represented the vertical assistant bet. Pi was designed as a personal AI with proprietary models and a distinct conversational product surface. That created a tighter user experience, but it also tied the company much more closely to winning on one assistant experience instead of becoming infrastructure for many agent workflows.
  4 sacra 5 techcrunch 6 time
• This makes /dev/agents a different kind of wager. The company is closer to an operating layer that coordinates context and actions across devices, similar to other OS level approaches emerging on Android, where deeper data access can make agents more useful but distribution becomes much harder.
  7 sacra 8 sacra

Going forward, the market is likely to separate into three durable layers, assistant brands, workflow agents inside existing software, and OS level coordination. The biggest upside sits with whoever controls cross app context and action routing, because that layer can decide when a standalone assistant is enough and when a specialized agent should take over.