If you only have a few minutes to spare, here’s what investors, operators, and founders should know about StackAI (W23).
StackAI built a no-code layer for connecting large language models to the systems where enterprises keep data and execute work. Founded in 2023 by MIT PhDs Toni Rosinol and Bernard Aceituno, the YC Winter 2023 company evolved from a developer API into a visual workflow builder with more than 100 integrations, private deployment options, and controls for regulated buyers.[1][2]
Its outcome was an acquisition, not a collapse. Asana bought StackAI in May 2026 for about $75 million in cash, subject to adjustments, while offering employees equity awards and an earnout opportunity.[3][4] The deal compressed two complementary assets: StackAI had built cross-system execution and enterprise controls; Asana already had collaborative context, approvals, distribution, and thousands of customers. The same complementarity explains both StackAI's value and the difficulty of scaling it independently as a horizontal platform.
Rosinol and Aceituno began working on the company in 2022, before incorporating the business that YC lists as founded in 2023. Both held MIT doctorates. Public sources do not establish how they met, so any more specific origin story would be conjecture. Their initial problem was concrete: enterprises wanted to use language models, but useful work required connecting those models to private data and operating systems.[1][2]
The first product was an API for developers. Customer work on requests for proposals and sales automation changed the founders' view of the bottleneck. Training another model was less important than reliably moving context into a model and approved actions back into business software. By May 2024, StackAI had become a low-code canvas where users arranged data sources, models, triggers, and actions into workflows.[5]
That pivot moved StackAI toward business teams while increasing the enterprise burden around security, permissions, deployment, evaluation, and implementation. It also created the asset Asana later wanted. A workflow agent that only drafts text is easy to copy. One that can read from SharePoint, reason over governed data, request an approval, and write into an ERP is much harder to ship safely.
Aceituno summarized the early insight to TechCrunch: “Our platform allows people to build workflows that require connecting different tools to work together.”[5] At acquisition, Rosinol drew the sharper product boundary: “General-purpose agents talk; specialized agents act.”[6] The first quote explains the original product; the second explains why Asana bought it.
StackAI gave teams a drag-and-drop canvas for assembling an agentic workflow. A user could connect a source such as SharePoint, SAP, Workday, Salesforce, Snowflake, or Miro; add a model; supply governed context through a knowledge-base node; expose approved tools; and define what happened next. The workflow could read, write, or execute across enterprise systems rather than merely produce an answer in a chat window.[7]
The mature product separated the model from the operating layer. Customers could select providers including OpenAI, DeepSeek, and Perplexity while StackAI handled connections, retrieval, tools, deployment, and monitoring. Pre-built interfaces let teams deliver workflows without building a custom front end. This mattered because an enterprise agent is rarely one model call. It is a chain of access checks, retrieval steps, model decisions, human approvals, and writes into systems of record.
Read the complete post-mortem, the rebuild playbook, and the exact reasons StackAI is still worth studying now.
