Interview
Rebuilding IT From the Ground Up for the AI Age: Serval's Jake Stauch
- The company aims to close the gap between job visions and reality by automating repetitive tasks, acknowledging that this outcome is not certain for every profession, though it remains the core mission.
- Traditional workflow implementations, which can take weeks to months, may lag behind rapidly changing business needs, whereas the AI-native approach targets practically zero time to develop.
- To prevent workflow duplication and AI confusion, the company plans to deploy an agent with full contextual awareness to identify, modify, or delete existing similar automations.
- Future competitive advantage is anticipated to rely on customer insight and empathy rather than product features, which are easily copied overnight.
- Product longevity will be ensured by building on boundaries such as permissions, approvals, audits, and logs to allow enterprise use of powerful models without elevating security risk.
- A two-pronged architecture will be implemented where an admin agent controls tools and skills, while a help desk agent operates with full intelligence only within an approved scope.
- Current model strategy involves using OpenAI models for end-user interactions and Anthropic models (specifically Sonnet and Opus) for automation code generation, though these selections may shift as new models are released.
- Integrating new models into production is not fully automated and currently requires manual prompt tuning and infrastructure adjustments for known quirks.
- While new models may be slightly smarter, their increased unpredictability may lead to downgrades to older, more reliable models.
- Costs are not the current primary focus, but the company expects expenses to become critical as it explores background and long-running agents, necessitating earlier monitoring to prevent costs from running away.
- The company anticipates that large foundation model labs will not prioritize the complexities of enterprise service management, viewing it as a poor resource allocation.
- Large enterprises face adoption challenges due to committee structures and coordination issues, which serve as the rate-limiting step compared to smaller AI-native companies.
- The product is expected to change employee experiences in large organizations by preventing employees from being blocked for weeks while awaiting workflow implementation.
- Hiring remains a primary concern with a mantra of "fewer, better" to maintain talent density, as rapid scaling could hinder the company's ability to reinvent itself.
- The organization expects to reinvent itself more frequently than traditional companies, including renaming product parts and shifting direction almost overnight.
- Companies that default to "yes" for AI adoption will be ahead of those that default to "no," despite facing potential security incidents and consequences.
- The long-term impact goal is for the company to be recognized for making work lives better by unlocking meaningful work rather than solely automating IT jobs.
- The corporate culture is not designed for training, mentorship, or those seeking clearly defined career paths, as the organization aims to remain as flat as possible.