Despite industry talk, there is currently no software that can orchestrate and manage various third-party AI agents from different vendors. Teams must manage each agent in its own siloed interface, creating significant operational overhead.

Lab work is "high mix, low volume," like driving, making it hard to automate. Traditional automation is like a subway: efficient but inflexible. AI enables "autonomous" labs, akin to Waymo cars, that handle the vast variability of experiments, which constitutes 99% of lab work.

The usefulness of AI agents is severely hampered because most web services lack robust, accessible APIs. This forces agents to rely on unstable methods like web scraping, which are easily blocked, limiting their reliability and potential integration into complex workflows.

The next leap for hardware—AI generating complex 3D CAD designs—is blocked by a data bottleneck. CAD files are a company's most valuable IP, so firms won't share them to train models. The solution may lie in on-premise models or starting with the hobbyist community.

Scientists won't adopt automation if they have to code or use clunky visual programmers. The breakthrough is using AI models to translate natural language protocols into robot commands. This removes the primary usability barrier and prevents common user errors, enabling adoption.

To create a complex automated science platform, first build modular tools that human experts use in a manual workflow. Observe their process to identify bottlenecks and needed components (e.g., a stability test). Then, incrementally build agents to automate the orchestration of these proven tools.

The primary value of AI in bioprocessing is not just automating tasks, but analyzing process data to predict outcomes. This requires a fundamental shift in capital equipment design, focusing on integrating more sensors and methods to collect far more granular data than is standard today.

When selecting new software, the primary evaluation criteria should be its potential for integration with AI agents. Look first for a Command Line Interface (CLI), then a platform connection like an MCP, and finally, a robust API. This prioritizes automation capability over user-facing features.

Automating science involves solving mundane physical problems. Radical AI had to design custom actuators just to unstick material samples from trays—a task a human does intuitively with a chisel, highlighting the often-overlooked 'last-mile' challenges in robotics.