Maxime Beauchemin and the trouble with doing things twice

A three-day hackathon is an odd place to begin a business-intelligence project. The calendar suggests a small experiment, preferably one that can be demonstrated before everyone goes home. At Airbnb in 2015, Maxime Beauchemin had a practical irritation to work on: the company was investing in databases that its visualization tools did not comfortably serve. Druid could answer questions quickly. Getting those answers onto a useful screen was another matter.

He built a front end. That project became Apache Superset. In the preceding year, he had already started the workflow engine that became Apache Airflow. Two different problems, two pieces of software that would travel well beyond the company where they began. There is something pleasingly disproportionate about the sequence: an engineer tries to make his working day less awkward, and other people inherit the improvement.

Today Beauchemin is the founder and CEO of Preset, the company he started in 2019 around Superset. His career makes an unusually coherent argument for paying attention to friction. The small inconvenience at your desk may be a shared problem. Fixing it properly can turn into a rather larger job than you bargained for.

Before the dashboards, the warehouse

His working vocabulary was assembled across several kinds of company. At Ubisoft, he worked as a data warehouse architect. At Yahoo, he worked on clickstream analytics and adopted Hadoop and Pig early. At Facebook, his work included frameworks for engagement and growth analytics. These were different businesses with a familiar requirement: turn accumulating records into something people could use.

The progression matters because data engineering has never been only a matter of choosing a database. Someone has to collect information, arrange it, move it, and keep the machinery running while colleagues ask new questions. Beauchemin encountered that work before it acquired the contemporary assortment of specialist titles.

In January 2017, he published “The Rise of the Data Engineer.” He described a discipline bringing software-engineering methods to business intelligence and data warehousing. Code, reusable infrastructure, and automation were becoming central to a job that had previously been associated with proprietary tools and report production. The essay offered practitioners a vocabulary for work they were already doing. A job description can be useful infrastructure too.

A calendar with dependencies

Airflow started at Airbnb in October 2014. Its central concern was sequence: one task needs another task to finish before it can begin. A collection of scripts becomes troublesome when those relationships are implicit, scattered, or known mainly to the colleague who wrote them. The software made workflows something engineers could define as code, schedule, and inspect.

Beauchemin had seen an internal Python-based system called Dataswarm at Facebook. When considering Airbnb, he discussed building something that drew on those ideas and could be shared publicly. His previous tools and practical experience supplied other ingredients. Airflow grew from that mixture, rather than from a sudden discovery that computers could run tasks on a timetable.

It was open source from its first commit, formally announced under Airbnb’s GitHub organization in June 2015, and entered Apache incubation in March 2016. By January 2019, it was an Apache top-level project. Those dates describe a change in custody as much as a release schedule: an internal solution acquired a community and a governance structure beyond its original employer.

His June 2015 introduction emphasized the ability to author, monitor, and retrofit pipelines while the company grew. Retrofitting is the revealing word. Real systems arrive with yesterday’s decisions attached. An orchestration tool earns its keep among those decisions, where the diagram is untidy and the next deadline is already approaching.

Three days to get a picture

Superset addressed the other end of the journey. Once data existed, people needed to explore it. In Beauchemin’s account of Airbnb’s setup, the company had invested in Presto and Druid, while its existing visualization arrangements introduced limitations around connectivity and extracts. Druid, at the time, did not speak SQL. A quick interface for exploring its datasets offered a concrete starting point.

The application was once called Panoramix; by a March 2016 interview, Beauchemin was discussing it alongside Airflow. The eventual Superset would combine visualization, dashboards, and a SQL editor. Its audience extended beyond the people maintaining the pipelines. An open-source application with an interface for analysts presents a different challenge from a library that programmers call inside their own code.

There was a personal attraction here. Beauchemin also made digital art, including colorful, interactive work associated with Burning Man. Visualization let him work on something people could see and manipulate. He has described that visual dimension as part of what drew him toward Superset. The charts were useful, and the work itself interested him.

Superset entered Apache incubation in May 2017. In January 2021, Apache announced its top-level status, with users including Airbnb, American Express, Dropbox, Lyft, and Netflix. A hackathon supplies the opening scene; years of contributions, releases, and community work explain how the application stays on screen.

The author who punctured his own balloon

Seven months after “The Rise of the Data Engineer,” Beauchemin published “The Downfall of the Data Engineer.” The titles look like an argument conducted by one person who refuses to wait for a reply. He acknowledged the sensational framing and explained that he still believed in the discipline. His subject was the difficulty of doing the work well.

He catalogued context switching, maintenance burdens, tangled dependencies, and the way infrastructure work receives attention when something breaks. A team can become so occupied supporting what it built that it has little time left to improve it. More machinery can help engineers keep additional tasks moving while leaving the underlying conditions frustratingly familiar.

This was a useful complication to the creator’s success story. Beauchemin had helped build a tool for managing pipelines and was willing to describe the limits of better pipeline tooling. His criticism concerned incentives and working arrangements as well as software. The practical question was how engineers could regain room to build.

In 2018, “Functional Data Engineering” continued that effort through an argument for more reproducible batch processing. Across these pieces, he kept returning to how work is expressed and maintained. The professional title was worth explaining; the daily experience deserved equal scrutiny.

“The goal is to automate most of the automatable CEO-stuff”Maxime Beauchemin, in conversation with Simon Späti, 2026

A company around the commons

After Airbnb, Beauchemin worked at Lyft, developing open-source products for working with data. In 2019 he founded Preset. The company put a commercial service around Superset: customers could use the analytics software while paying someone else to handle the operation and supporting capabilities. Public code and paid convenience could coexist.

In August 2021, Preset announced a $35.9 million Series B led by Redpoint Ventures, with participation from existing investor Andreessen Horowitz. That round brought the company’s announced funding to $48.4 million and coincided with the general availability of Preset Cloud. Tomasz Tunguz joined its board. The transaction funded a business built around making an existing community project easier to adopt.

Beauchemin’s argument for open source reaches beyond price. In his 2021 essay on the future of business intelligence, he emphasized discovery, experimentation, collaboration, and the freedom to operate software in different ways. People could inspect and extend the system they used. For an engineer accustomed to integration problems, that freedom had a practical value.

There are two institutions in this story. Apache Superset has community governance; Preset sells services and products built around it. Keeping those roles legible helps explain Beauchemin’s move from engineer to founder. Building software that others can adopt and building a company they can buy from involve overlapping, distinctly demanding responsibilities.

The next colleague may need permissions

By 2024, Beauchemin was urging engineers to develop a reflex for trying AI assistance in their work. His discussion with Software Misadventures focused on the value of business context and judgment as routine SQL generation became easier. He also discussed Promptimize, a toolkit applying testing ideas to prompts. Experimentation came with a way to compare results.

The same year’s enthusiasm has since acquired an organizational shape. In October 2025, he introduced Agor, a spatial, collaborative environment for agent-assisted development. Its premise was familiar: faster execution creates coordination problems elsewhere. Branches, environments, reviews, and conversations still need to meet somewhere. A colleague’s useful work can remain surprisingly inaccessible inside that colleague’s terminal.

In March 2026, he introduced persistent Agor teammates with identity, memory, skills, and scheduled activity. Later that month, he proposed the role “AI Enablement Engineer” for people who help an entire organization use the tools. The emphasis moved from an individual’s productivity toward shared access, instructions, and working practices.

His July 2026 conversation with Simon Späti explored a more visual way to coordinate agents and the permissions those agents need. One ambition was to automate routine parts of his CEO work. Yet the organizational question survives automation: who can see the work, understand it, and decide whether to accept it? A faster colleague still needs an intelligible job.

Another side project leaves the desk

Preset announced a $7.27 million Series C on March 9, 2026, led by Andreessen Horowitz with Redpoint and Datatech.fund participating. Beauchemin reported more than 400 customers. He described investment in AI capabilities and the Superset ecosystem, extending the company’s work from analytics for people toward analytics that agents can also use.

On October 13, Preset announced Agor Cloud’s open beta. Beauchemin traced the product to his own difficulty keeping track of a growing number of agents, then to the larger problem of making their work useful across a team. The hosted service takes on infrastructure work for customers. That progression bears a family resemblance to Preset’s original offer around Superset.

The announcement also supplied a pleasantly unpolished example. An internal competitive-intelligence agent called Wendy had mistaken a pricing-page experiment for a real pricing change. The correction became a recorded lesson. Beauchemin’s account made review and accumulated context part of the product’s working life. Even a synthetic colleague can misunderstand the office gossip.

Read across the projects and a consistent preference emerges: make the process visible enough for other people to participate. Pipelines, charts, and shared agent boards give different work a shape that colleagues can inspect. Beauchemin keeps finding fresh complications in that proposition. There is, apparently, plenty left to do before everyone can stop doing things twice.