A bit of backstory: we started out doing white-glove implementations of agents/workflow automation for large enterprises (consumer goods, financial services). First customer was Red Bull which helped us get to profitability and remain bootstrapped. A few things became clear along the way:
- The biggest adoption blocker was never the tech, it was behavioural: customers didn’t want months of setup and demos/trials mostly because: (1) models + tooling were evolving very fast (decision paralysis) and (2) a lot of enterprise tools required big up-front commitments with no guarantee on uptake/usage. So almost all our large customers started with one use-case, tracked usage, then expanded scope and adoption from there. The question for us became how to create that journey with minimal handholding. We’re now pushing a model of start free -> see results -> scale with usage to expand scope to smaller businesses too.
- Users didn't want a new UI. Usage grew when they could interact with agents from existing surfaces (email, Slack, WhatsApp - whatever fits their style), as it minimised the behavioural friction.
- Grounding in internal data was really important so we spent a lot of time tuning document indexing + retrieval.
- The bespoke, service-heavy model that drove our early growth was sticky but maintenance was a pain given the pace of change (especially as a lean bootstrapped company). Given that many customer were reusing variants of the same underlying primitives + agent harness, we focused on finding ways for customers to update their own agents/workflows and knowledge configurations with minimum friction.
We're still mid-market/enterprise by background and new to self-serve so if something's confusing or clunky let me know.
A bit of backstory: we started out doing white-glove implementations of agents/workflow automation for large enterprises (consumer goods, financial services). First customer was Red Bull which helped us get to profitability and remain bootstrapped. A few things became clear along the way:
- The biggest adoption blocker was never the tech, it was behavioural: customers didn’t want months of setup and demos/trials mostly because: (1) models + tooling were evolving very fast (decision paralysis) and (2) a lot of enterprise tools required big up-front commitments with no guarantee on uptake/usage. So almost all our large customers started with one use-case, tracked usage, then expanded scope and adoption from there. The question for us became how to create that journey with minimal handholding. We’re now pushing a model of start free -> see results -> scale with usage to expand scope to smaller businesses too.
- Users didn't want a new UI. Usage grew when they could interact with agents from existing surfaces (email, Slack, WhatsApp - whatever fits their style), as it minimised the behavioural friction.
- Grounding in internal data was really important so we spent a lot of time tuning document indexing + retrieval.
- The bespoke, service-heavy model that drove our early growth was sticky but maintenance was a pain given the pace of change (especially as a lean bootstrapped company). Given that many customer were reusing variants of the same underlying primitives + agent harness, we focused on finding ways for customers to update their own agents/workflows and knowledge configurations with minimum friction.
We're still mid-market/enterprise by background and new to self-serve so if something's confusing or clunky let me know.
Jai
Kushal here from the engineering team. Happy to answer any technical questions about our approach or how we built the product.