The useful thing about a robot failure is that it leaves clues. A camera saw one scene. A lidar traced another. Motors received commands, software changed state, and a clock tried to keep the whole episode in order. Somewhere inside those streams is the reason a vehicle braked, a drone drifted, or an arm missed its grip. The hard part is getting a human being to the right few seconds.
Adrian Macneil has built a company around that search. Foxglove, which he co-founded with Roman Shtylman in 2021, gives robotics teams a place to record, manage, inspect, and replay multimodal data. Its interface can put maps, video, point clouds, plots, and state messages on the same screen. The product is technical, but the promise is easy to understand: when a machine does something strange, the people responsible should be able to see why.
Macneil arrived at this problem by moving through several industries just as their infrastructure was being invented. He built e-commerce and payment integrations, joined Coinbase soon after moving from New Zealand to the Bay Area, and then spent five years leading infrastructure engineering at Cruise. Each stop rewarded the same instinct: look past the visible product and study the machinery that lets a team build it repeatedly.
“Don't start a startup unless it is something that you're ready to work on for 10 years.”Adrian Macneil, Day One podcast, 2025
The orchard is still in the picture
Macneil grew up in a rural, fruit-growing part of New Zealand. His high-school jobs included pruning and packing kiwifruit. He has described sitting beside a conveyor, watching fruit pass, spotting blemishes, and moving pieces into different grades. It was repetitive work with a clear rule and a physical consequence. Years later, he would point to that memory when talking about tasks that no longer seem sensible for people to perform by hand.
The anecdote keeps his robotics thesis grounded. Labor productivity can sound like an economist's abstraction. In a packhouse, it is a person watching the same motion all day. Macneil is interested in the ten- and twenty-year horizon of technology, but his examples tend to be concrete: forklifts, tractors, warehouse machines, fruit. He likes futuristic systems when they are close enough to become useful.
He studied computer science at Massey University and worked as a software engineer at the New Zealand Treasury. Then came e-commerce. He created Exp:resso, a platform for the ExpressionEngine publishing system, and built payment gateway integrations. The company was acquired in 2014. That work also nudged him toward Bitcoin. His first reaction was skepticism; the slow and complicated reality of conventional cross-border payments made him reconsider.
Software engineering at the New Zealand Treasury.
Exp:resso is acquired; Macneil moves to the Bay Area and joins Coinbase.
Begins leading infrastructure engineering at Cruise.
Co-founds Foxglove with Roman Shtylman.
Foxglove announces a $40 million Series B.
A car with no left turn
Coinbase hired Macneil in 2014 as a full-stack software engineer and introduced him with a joke about fielding questions on Hobbits and Flight of the Conchords. He became the company's first Director of Engineering and helped build its digital currency exchange. It was a front-row education in what happens when a young technical market suddenly needs dependable infrastructure.
Cruise supplied the physical version of that lesson. Macneil remembers his first ride in 2016: the car could not make left turns and braked sharply even in an empty lane. By 2022, after he had left the company, he hailed a fully driverless Cruise and found the experience both remarkable and quickly ordinary. Once he looked at his phone, the empty front seat receded from attention. The product had crossed an experiential threshold.
Behind that improvement sat a much larger engineering organization. When Macneil joined, Cruise was still a small company working from early prototypes. He watched it grow to well over a thousand employees. A robotics program at that scale needs far more than autonomy algorithms. It needs builds, fleet systems, data pipelines, recording, deterministic replay, and ways for hundreds of engineers to inspect the same event.
One crucial visualization system began during a Cruise hackathon in 2017. Engineers made a browser-based proof of concept because the existing workflow could be better. Macneil is careful about the credit: he led the organization, while the engineers built the tool. That distinction reveals something useful about his operating style. He tells the story as an example of capable people recognizing a pain before it became an executive mandate.
The browser mattered. A specialist setup on one engineer's machine can work for a tiny team. At scale, a URL is more powerful. It works across operating systems, reaches remote data, and makes collaboration feel like the rest of the web. A colleague can open the same recording instead of walking over to stare at someone else's screen.
Building the layer everyone was rebuilding
Macneil kept seeing autonomous-vehicle companies reproduce the same tooling. The richest teams could spend heavily on bespoke infrastructure. Smaller robotics companies needed many of the same capabilities but could not justify the same investment. Foxglove emerged from that mismatch. The company would not build a robot. It would build the common layer beneath many kinds of robots.
The data is unusually unruly. A conventional observability tool is comfortable with text logs and time-series metrics. A robot can produce video, audio, 3D point clouds, positions, joint states, and messages at different frequencies, often with poor connectivity at the edge. Engineers need to preserve timing, move large recordings, search for rare events, and replay what the machine perceived alongside what it did.
Foxglove's answer has expanded beyond the original visualization studio. It includes data recording and ingestion, indexing and search, device management, collaboration, APIs, and cloud or self-hosted storage. The company also helped create MCAP, an open container format designed for multimodal robotics data, and joined the ROS 2 Technical Steering Committee in 2022.
The surprising part is how horizontal the workflow becomes. A drone and an autonomous boat do not move alike. A warehouse arm and a tractor have different sensors, environments, and safety constraints. Yet their developers still ask a common set of questions: What happened? When? What did the machine perceive? Can another engineer reproduce it? The interfaces differ; the investigation rhymes.
A flower beside the road
For a company built around complicated data, the name has a pleasingly simple origin. During a road trip through New Zealand with his wife and child, Macneil noticed foxglove flowers along the roadside. He liked the word. It was pronounceable, spellable, and available. There was no hidden robotics reference.
That preference for legibility appears elsewhere. Macneil has argued that people from ordinary software backgrounds should not count themselves out of robotics. A modern robotics team also needs front-end developers, infrastructure engineers, data specialists, and people who understand deployment. Not everyone has to arrive with a doctorate in machine learning or control theory. The field becomes more capable when those disciplines meet.
In November 2025, Foxglove announced a $40 million Series B led by Bessemer Venture Partners, with Eclipse, Amplify, and Icehouse Ventures participating. Macneil framed the money around a larger transition: sensor technology, edge computing, and foundation models were making “physical AI” commercially relevant, while the data infrastructure underneath remained incomplete.
The phrase can invite grand predictions. Macneil's more persuasive argument is operational. Starting a robotics company should eventually feel less like constructing an entire computing stack and more like starting a software company with cloud services, frameworks, and standard formats already waiting. Founders should be able to focus on the customer and the machine instead of rebuilding every layer beneath them.
“We believe that robotics will have a massive positive impact over the next decade, and our mission is to build the platform to make it possible.”Adrian Macneil, 2025
This is why the ten-year test matters. Foxglove began when robotics infrastructure was still a specialized pitch. Five years later, the company said it served more than 250 customers across autonomous vehicles, logistics, agriculture, defense, marine systems, and other categories. The robots vary. The need to understand them does not.
Macneil's career has moved from digital payments to autonomous motion, but the recurring subject is trust in complex systems. Money has to arrive where software says it went. A vehicle has to behave as its models intended. A robot failure has to become an explanation a team can act on. The glamorous moment is the machine moving by itself. The enduring work is making that moment observable, repeatable, and eventually ordinary.