We sometimes share strategy updates with customers. It gives us a reason to write down what we're seeing, let people push on it, and show more of how we're thinking than normally makes it into product updates.
Delegation has been a topic for a while. Honestly, I think it is getting pretty close to consumer-ready. Models have moved from answering questions, to using tools in a thread, to coordinating other agents. That moves automation from individual jobs toward whole functions, and creates a much bigger question for companies than which chat tool they should buy.
We put together packets like this with customers in the spirit of openness. This one is about what delegation changes inside a business and in the market, why reusable company functions become more valuable, and how we think qbash Tasks fit underneath people, systems and agents.
Figured I'd share it here as well.
Read the deck as text
The next capability jump is delegation
AI systems have moved from answering questions, to chaining prompts, to using tools and subtasks inside a thread. The next step is coordination: creating teammates, passing context between them and responding to their results. Automation moves from completing an individual job to running more of a whole function.
More of a business becomes reproducible
Businesses sell more than isolated outputs. They organise the work behind those outputs. As delegation improves, AI can reproduce more of that coordination and delivery, including research, reporting, filing, monitoring and other repeatable service work.
Customers are spreading across more tools
Companies once made a fairly simple choice between a few general assistants. People now work across specialised products, coding tools, voice tools and agent runtimes. The agent choosing a tool may never expose the company behind it. API or MCP access makes a business callable, but does not guarantee distribution.
Businesses designed as functions have an advantage
Companies that define their work as reusable functions can carry those operations into each new capability jump. Agents can call the same functions at greater volume. Processes trapped in chat history, prompts or one agent's context need to be rebuilt whenever the surrounding tool changes.
The company needs builders
Leadership needs to understand what is becoming possible, domain experts need to define the work, and technical builders need to turn that knowledge into reliable operations. The people best equipped to run today's company may not be the people who can build its next version.
What we expect next
The work AI absorbs will keep moving upstream into coordination and delivery. Competition over delegation will be intense. Generic chat layers will matter less as major products absorb model access, tools and basic coordination. At the same time, customers will use a wider mix of specialised products, creating more points of use and distribution.
The qbash toolkit
qbash combines Tasks, connectors, MCP servers and Memories. Tasks contain the application logic. Connectors provide authenticated access to business systems. MCP servers publish selected Tasks, connector actions and Memories to people or agents. Memories keep documents and findings that applications can search and reuse.
Tasks are AI-enabled applications
A Task is a Python application hosted by qbash. Code defines the repeatable process and calls AI where judgment is needed. A Task can also wrap a service as a controlled tool for a person or agent. The SDK includes model calls, agent runtimes, Memories and connectors, with versions and logs around the finished application.
Tasks, agents and canvases
Agents are useful when the next step is uncertain. Tasks are useful when the rules are known and the work should run repeatedly. A Task keeps the process in code, with model use limited to the steps that need judgment. qbash handles execution, connectors, credentials, revisions and run logs around it.
One layer for people, systems and agents
The same Task can be used from a chat tool, run on a schedule or system trigger, or given to an agent as a tool. qbash manages the shared execution layer independently of the chats, systems and agents using it. The deck closes with examples across advertising, engineering, sales, support and recurring reporting.