When we announced that Aisle was becoming qbash, I wrote about how the product had changed underneath the old name. We started with prompts and a visual workflow engine, then watched the useful work turn into applications: systems that replace a unit of work in the business, often without an end user involved at all.
The rebrand was part of a larger repositioning around those observations. Frankly, we started this journey two and a half years ago, and anybody who works with me knows one of my bigger regrets is that we weren't publishing content sooner. In that spirit, I thought it would be interesting to share our strategy deck about the reposition and how we see ourselves fitting in. It goes further into what succeeded, what we shut down, the difference between end-user tools and background systems, and why we think this new class of application needs its own layer.
Read the deck as text
History
We made three bets. First, reliable AI comes from chaining bounded prompts into a deterministic process, rather than asking one large prompt to do everything. That has become industry consensus. Second, we gave end users a visual canvas for constructing their own solutions. In practice, the work stayed trivial, and an agent can now cover much of it. Third, we expected companies to need an internal portal for distributing AI applications. As models commoditized, the largest providers built their own distribution and lock-in through the user experience.
What persisted
One type of implementation stayed critical through every platform change. We distinguish between tools that an end user builds for themselves and systems that replace units of work in a business. The systems run hundreds of times a month, may never engage an end user at all, and keep running because operations depend on them.
The iceberg
ChatGPT, Claude, Copilot and Gemini are the visible tip. Below the waterline are the systems doing background business work: ingesting documents, deduplicating and routing leads, detecting vendor price changes, and hundreds of other small pieces of operational software.
Why Tasks succeeded
We build these systems as Tasks: the process in deterministic Python, with judgment delegated to bounded model calls. This makes spend predictable, keeps the company’s process and intellectual property outside any one model, allows different models to be chosen for different steps, and leaves API calls under the control of code. The result also has the ordinary properties of good software: repeatable execution, versions, tests and logs.
Tasks, agents and canvases
An agent keeps the process in the model’s head. A workflow canvas stores it as a proprietary node graph. Building from scratch gives you code, but also makes you responsible for the runtime around it. A qbash Task is Python with the framework in scope: diffable, testable, reviewable and readable by the next developer or model.
The code, and where it runs
A Task can call normal integrations directly, then use a model for a bounded judgment step. Because the model, integrations and launch surface are all treated as interfaces, the same Task can survive a change in provider and run from Claude, ChatGPT, Copilot, Gemini, Cursor, Slack, Gmail, Jira, an API or a schedule.
What we see today
Programmatic Tasks now dwarf the other usage on the platform and are growing faster than anything we have shipped. They spread through an organization: one builder ships a system and the rest of the organization becomes users of it. Agencies are also adopting them because building and maintaining these systems on qbash is easier than running the surrounding infrastructure themselves.
The market thesis
These systems are replacing some applications that would once have been built conventionally, plus a much larger set that was never built because it was too expensive. The growth is visible in the infrastructure around them: DigitalOcean reported 29% year-over-year revenue growth and 212% growth in AI customer ARR in Q2 2026; Anthropic said its run-rate revenue crossed $47 billion in May 2026; its September 2025 Economic Index found that 77% of business API usage showed automation patterns; and n8n was valued at $5.2 billion in May 2026.
The labs are incentivized to own the interface, the clouds to sell compute, and the workflow canvases to preserve the node graph. We think this class of application is becoming a category of its own, and intend qbash to be the best place to build it.