Decagon found that fine-tuning smaller, specialized open-source models led to better performance, lower latency, and reduced costs compared to using general-purpose frontier models for specific enterprise workflows. This contradicts the common "smart but expensive vs. dumb but cheap" tradeoff.
Decagon uses powerful frontier models not for high-volume core operations, but for complex, open-ended auxiliary tasks like system-wide performance analysis and model improvement. This hybrid approach optimizes for both efficiency in core tasks and intelligence in strategic ones.
While model performance is key, the real defensibility for enterprise AI applications lies in the surrounding software stack. This includes tooling for compliance, testing, integrations, and business logic management, which are necessary to make powerful AI safely deployable within large organizations.
The "forward deployed engineer" (FDE) role is a temporary bridge for AI startups to discover customer workflows. If learnings aren't rapidly productized into the core platform, the company risks becoming an unscalable consulting business, not a tech company. The FDE's output must feed the core product.
Decagon developed "Duet," a secondary AI agent that handles the entire lifecycle of their primary customer-facing agents. It automates writing operating procedures, generating tests, and monitoring performance, demonstrating how AI can be used to manage the complexity of building production AI systems.
Contrary to the popular focus on "tokenomics," Decagon states that for a growth company, optimizing for performance and latency is paramount. Their shift to a cheaper open-source stack was driven by a need for speed, with cost savings being a secondary benefit, not the primary objective.
Decagon won a key customer from a competitor by providing a productized platform (a "glass box") that empowers the customer's team to build and iterate on their own. This contrasts with the competitor's "black box" service model, which relied heavily on forward deployed engineers and created a bottleneck.
Decagon accelerates enterprise sales cycles by providing customers a detailed roadmap for adoption. This includes navigating internal processes like model risk governance and security reviews. For large enterprises, understanding *how* to deploy AI safely is as important as what the AI does.
Instead of simply replacing human agents, deploying efficient AI customer support often leads to a surge in demand. Companies make support more accessible (e.g., putting it on every page for free users), which increases overall consumption, a classic example of the Jevons paradox.
As the model landscape changes rapidly, AI application companies must operate an internal "model factory." Decagon Labs continuously fine-tunes new open-source models for their specific use cases, creating a system to quickly leverage advancements and maintain a performance edge.
Even super-intelligent AI agents will require software like databases and CRMs as systems of record. Similarly, human careers will persist post-AGI because most jobs are abstract constructs designed to serve other humans, a need that won't disappear. The nature of work will change, but not vanish.
Unlike traditional SaaS, AI startups are expanding internationally at a much earlier stage. This is driven by universal top-down pressure on enterprises to adopt AI and the relative ease of localizing language models, leading to strong customer pull from different geos.