Interview, Fireside Chat, Conference Presentation
The Future of Digital Workers
- The company aims to utilize agents only when necessary to achieve customer outcomes, planning to re-architect the product to be "super agentic" with the expectation of delivering superior results within three months compared to the current platform.
- Prospective customers will require training regarding the non-deterministic nature of agent outputs, as the speaker anticipates the term "agent" will be widely misused despite true agents being defined by their ability to plan, reason, reflect, and improve over time.
- B2B verticalized agentic applications are expected to be more complex and difficult to solve than code generation due to multi-stakeholder involvement and subjective success criteria, necessitating a shift away from traditional SaaS feature requests to focus on the "digital worker" hypothesis.
- Specific product functions are projected to include Alice leveraging first-party CRM and third-party data warehouse data to target visitors and drive qualified meetings on autopilot, while Mike operates across multiple languages 24-7 for inbound and consented outbound use cases with deep CRM and calendaring integration.
- The technical strategy involves building complex orchestration for certain tasks while applying agentic processes to others like content writing and deep research, utilizing a tech stack designed to swap different models daily for tasks such as tool calling or instruction following.
- Significant advancements in voice technology are anticipated over the last two months and the next six months, moving toward voice-to-voice models that handle speech and inference within a single model to reduce latency, though the speaker expects most competitors to fail in achieving all three critical requirements of intelligence, natural speech, and accuracy simultaneously.
- Product development and scaling investments will be adjusted weekly, with small pods prioritizing validation over scale to reach Product-Market Fit quickly, while feature development will focus on making invisible agent actions visible to build customer trust.
- Customer education and confidence will be established through free access to new products during the sales process, using internal data metrics like email opens, clicks, and demo sign-ups to validate success across segments rather than relying on customer intuition.
- The company plans to leverage a "distributed army" of partner companies to test different approaches for rapid adoption, while maintaining a policy of saying "no" to features that dilute the product's agentic focus.
- Leadership and hiring strategies will emphasize the ability to move fast, deep ownership of customer success, and the capacity to handle chaos, including the willingness to kill work that no longer delivers outcomes, with the expectation that market conditions may require leadership to scale up or down instantly.
- The speaker expects the AI tooling landscape regarding agent frameworks, monitoring, evaluation, and multi-modal models to have changed significantly over the last eight months and to continue changing as significantly in the next six months, requiring human-in-the-loop feedback for large-scale model testing.