Summary
This episode focuses on how Decagon builds enterprise AI applications that automate customer support, sales, and operational workflows at scale. The founders argue that real enterprise value comes from productizing AI around business processes, not simply plugging into frontier models. They explain why Decagon moved most of its inference to open-source models, emphasizing latency, controllability, and task-specific performance over raw model intelligence. The conversation also covers the importance of forward-deployed engineers, deep evaluations, and workflow integration in production AI systems. Finally, they discuss the future of application-layer companies, the persistence of CRMs as systems of record, and why execution speed now matters more than model-side breakthroughs for commercial success.
Key Takeaways
- 1Enterprise AI products should be built around workflows and business processes, not just generating answers.
- 2Open-source models can outperform frontier models in production when latency, control, and task-specific quality matter most.
- 3Forward-deployed engineers are a strategic product function, not just a services layer.
- 4Enterprise AI needs a 'glass box' approach with transparency, governance, and deep evaluation.
- 5AI agents are likely to become the front door for business interactions, but traditional enterprise software will still matter underneath.
- 6Commercial advantage in enterprise AI is increasingly about execution speed and product quality, not just frontier model progress.
Notable Quotes
""The biggest breakthroughs in Enterprise AI aren't just happening inside Foundation models. They're happening in the products built around them.""
""So today, 90% of our workflow is on open source.""
""On the specific tasks we want them to do, they actually outperform the large smart state of the optimal.""
""I think there will always be a space for application layer companies.""
"But from a business perspective it's less on the model side and more on the just like can you build the company fast."
"So I think that's that's like a good, good way to do it."
"We found that in a, for a lot of our customers, there's actually just more demand for things like customer support than they're supply."
"AI will, uh, kill jobs, but not careers in a way, because like those jobs that are being done currently should not be done by humans."
Episode questions
Why did Decagon move so much of its inference to open-source models?
Because latency, control, and task-specific performance became more important as the company scaled. Jesse explains that smaller open-source models can be tuned to outperform larger models on the exact tasks Decagon needs.
What makes enterprise AI deployment harder than simply using frontier models?
Enterprises need benchmarks, evaluations, workflow controls, governance, and integration with legacy systems. Decagon says public evals are not enough and that companies need their own task-specific testing and deployment process.
Why does Decagon think application companies will still matter if AGI arrives?
The speakers argue that AI still needs a place to store work, pull information, and follow business processes. Even with very capable models, companies will need software layers that capture domain logic, compliance, and operational workflows.
How do Decagon’s forward-deployed teams differ from traditional services teams?
Their output is intended to become core product, not one-off customer work. The goal is to turn repeated customer pain into reusable features so the next customers get the benefit automatically.