A Year of Acceleration
2025 was a monumental year for AI. The list of major model releases, benchmarks, and product launches is long and well-documented elsewhere. Rather than recounting those developments, we wanted to share how the broader movement in AI shaped us as a startup—and a few things we consistently observed while working with customers across scientific and biomedical organizations.
Our Starting Hypothesis
Our hypothesis going into 2025 was straightforward: AI for science needs to be scalable, repeatable, and consistent. One of the core bottlenecks is the ability to use AI agents to reliably produce high-quality scientific assets. Everyday conversations with powerful chatbots have transformed our world, but this now-common modality often isn’t sufficient for scientific work. These one-off conversations don’t naturally compound to turn into durable datasets, reproducible analyses, or shared organizational knowledge.
From Conversations to Systems
Agentic pipelines, by contrast, can be designed to systematically generate clean, verifiable data assets. They can encode assumptions, track provenance, apply quality control, and be re-run at scale. The challenge—and the opportunity—is that moving from low-level agent frameworks to workflows that are genuinely useful in real scientific contexts is non-trivial.
Bridging this gap requires more than better prompts or more powerful models. It requires systems that make execution, verification, and reuse first-class concerns. This is the space we’ve focused on building into Devano, and where we consistently see the most durable value emerge for teams working with complex scientific data.
An Unexpected Shift in How We Work
What we didn’t anticipate was how deeply this same ethos would affect the way we worked ourselves.
As we built and deployed our agentic platform with customers, our internal bottleneck quietly moved. Writing code was no longer the limiting factor. Instead, the hard work became designing guidance systems, constraints, and quality-control procedures for the agents that now handle much of the coding and execution work we previously did by hand. In effect, as we helped customers build agentic automations for biomedical data, we were building and refining the same systems inside Devano—systems for guidance, verification, provenance, and scale. We were using the product the way it was meant to be used.
The impact on our productivity was dramatic. A team of two began operating with the leverage of a much larger organization, while making the services component of our enterprise business far more scalable, consistent, and repeatable.
Validation from Our Customers
Over the course of the year, this pattern was reinforced by our customers. Teams across scientific and biomedical organizations described the same limitations: chat-based AI interactions were useful and necessary, but often insufficient for building assets that could be trusted, reused, and scaled. What they needed were systems that could apply consistent logic, track provenance, and enforce quality standards across many datasets and use cases.
These conversations validated our direction. As customers incorporated Devano into their workflows, several described meaningful reductions in time spent on previously manual or ad-hoc tasks—often compressing work that took days or weeks into hours. Just as importantly, this shift has pushed teams to think more deliberately about how multiple AI systems fit together: where Devano provides structured execution and verification, and where other models and tools support exploration, review, or domain-specific reasoning. In practice, Devano has become one component of a broader, hybrid AI setup shaped by each team’s needs and constraints.
The Shape of Modern Work
In short, what we are seeing is not just AI adoption—and certainly not a world where a single AI platform replaces everything. Instead, we are seeing a fundamental and rapid shift in how work gets done: well-designed agentic systems plugging into an evolving, hybrid stack that augments human judgment. Increasingly, the human work is not in execution, but in the design of guidance systems and in figuring out how best to assemble the pieces.
Looking Ahead
Looking back on 2025, the biggest lesson for us wasn’t just about AI capability. It wasn’t even about biomedical data. It was about what the future of modern knowledge work looks like.
We thought we were building agentic pipelines for biomedical data. We didn’t realize we were on the bleeding edge of building the only way to work in the modern era.
In 2026, we’re continuing to build Devano around this belief: that the future of scientific work depends on systems that scale judgment, not just execution. If this way of working resonates with you, we’d love to compare notes—or help you build it inside your own organization.