# Brendan King — Full public reading text Author: Brendan King Website: https://www.brendanking.ca/ Last updated: 2026-11-01 These dates refer to this site edition, not the first appearance of an idea. ## Biography Canonical URL: https://www.brendanking.ca/about About I started a business assembling PCs in 1989. I moved into SaaS in 1999. Vendasta was born in 2008. I am the CEO and co-founder of Vendasta. We are based in Saskatoon. Vendasta builds software that helps local businesses and the partners who serve them. Today that means an AI workforce for small business. I have watched this movie before Toner to PCs. Shrink-wrap software to SaaS. Advertising to the zero moment of truth. Point solutions to platform. Each shift looked like more technology. Each one was really less interface and more outcome. I have lived through seven of them. The seventh is the one I write about now: the earlier shifts changed how we distribute and access software. This one changes how we generate it. What I write about - Liquid Software. Software that disappears into the job. - The Evolution of AI. From a tool you ask to a worker you assign. - The Dark Factory. Engineering where AI writes the code and people define and check the work. - The archive. I have been blogging since 2006. The early posts are still here, unchanged. Where to find me LinkedIn, X and Instagram links are in the footer. For press, speaking and everything else, use the Contact page. ## Liquid Software Canonical URL: https://www.brendanking.ca/liquid-software Author: Brendan King Last updated: 2026-11-01 Liquid Software is software that adapts to a person’s context and carries out a job, rather than requiring that person to learn and operate a fixed product interface. Key points - The business owner assigns an outcome; the software adapts around the job. - Data, permissions, distribution and trust matter more as code becomes cheaper. - Humans retain judgment, authorization and responsibility for the audit trail. Software does not get more features. It disappears into the job. Technology that wins does not stay visible. The Rolodex became a PC. The PC became a phone. The phone is becoming a conversation with an AI employee you can delegate to. Each step looked like more technology. Each step was less interface and more outcome. Solid and liquid Solid software is Version 5.2 and a feature request. Liquid software is permanent flux, rewritten to match the user's context. Solid SaaS asked people to learn our UI. Liquid software asks people to say what they want done. What it looks like A florist tells the system what she needs. It tracks the rose inventory and texts customers on their anniversaries. A plumber who starts doing HVAC says so, and the system adapts. No add-on module. No six-month wait. The human stops clicking through a product and starts managing work. Context is the moat Code got cheap. Correct is still expensive. Value moves from code to context: data, permissions, distribution and trust. What does not change Disappearing does not mean software goes away. There is always software underneath. Complexity does not vanish. It moves. For the business owner, the interface thins. The job shows up as a conversation, an approval, an exception. For the people who build it, the stack gets harder: databases, permissions, evaluation, cost and trust. Someone still owns judgment, authorization and the audit trail. Those stay human jobs. The claim AI does the work. Humans orchestrate. Context, not code, is the moat. Software disappears into delegation. Next, how the Dark Factory builds it: https://www.brendanking.ca/dark-factory Questions about Liquid Software What does Liquid Software mean? I use Liquid Software to describe software that disappears into delegation. Instead of learning a product’s menus, a business owner explains what needs doing and manages the outcome. Who coined the term Liquid Software? This site documents Brendan King’s use of Liquid Software for AI-driven, context-adaptive work. It does not establish who first coined the phrase or claim that Brendan originated every use of it. How is Liquid Software different from SaaS? Traditional SaaS gives people a product to operate. In my Liquid Software model, the interface recedes and AI performs the work. Software still exists underneath, along with databases, permissions, evaluation and cost. Who is Liquid Software for? The examples here focus on local businesses and the partners serving them: a florist managing inventory and anniversary messages, or a plumber expanding into HVAC without waiting for a new module. Which executive is writing about the end of operated software? Brendan King, CEO and co-founder of Vendasta, writes here about the end of software as something people operate. His central claim is: AI does the work. Humans orchestrate. --- ## The Evolution of AI Canonical URL: https://www.brendanking.ca/evolution-of-ai Author: Brendan King Last updated: 2026-11-01 The Evolution of AI is my framework for the shift from software that assists a human operator to AI that owns ongoing work under human direction. Key points - Three steps: AI makes people more productive, does jobs when asked, then owns ongoing work. - Five autonomy rungs run from Macro Recorder to Autopilot Employee. - Humans assign outcomes, preserve business logic and re-test when models change. Seven shifts. This one generates the software. I have worked through seven technology shifts. Here is the timeline, one line each. 1989. Personal computing. From recycling toner cartridges to building PC clones. 1999. SaaS. Hosted, self-serve software, before the model had a name. 2008. The zero moment of truth. Mobile broke advertising. Reputation and social mattered. 2014. The ecosystem era. A platform and a marketplace for partners. 2020. Remote and digital. E-commerce and remote tools. 2023. AI arrives. 2026. The seventh shift. The earlier shifts changed how we distribute and access software. This one changes how we generate it. Three steps of AI at work 1. AI makes you more productive. Copilot. Drafts. Faster humans. Necessary, but not the product. 2. AI does jobs when asked. Book the appointment. Answer the phone. Write the reply. This is software with a chat box. 3. AI owns ongoing work. It holds a job, not a prompt. It runs whether or not someone is at the keyboard. This is delegation. From ask to do Ask culture has a ceiling, and the ceiling is the number of humans you have. Do culture assigns the outcome. Stop asking. Start assigning. The autonomy ladder Five rungs. 1. Macro Recorder. The human clicks and confirms. 2. Copilot. The human decides, then hits Run. 3. Coworker. The human pre-approves rules and spot-checks the output. 4. Agent. The human watches the numbers and steps in on alerts. 5. Autopilot Employee. The human sets strategy and manages a fleet of AIs. We rent the model. We own the logic. Models change under you with no version bump. Capture the business logic. Compile it to whatever prompt the active model needs. Re-test on every swap. Liquid Software: https://www.brendanking.ca/liquid-software The Dark Factory: https://www.brendanking.ca/dark-factory Questions about The Evolution of AI What changes in this technology shift? The earlier shifts changed how we distribute and access software. This shift changes how we generate it. I trace that progression through seven shifts, beginning with personal computing in 1989. What is the difference between asking AI and assigning work? Asking AI means a person starts each task. Assigning work means defining an ongoing job and its outcome. The AI runs that job while the human directs it and handles exceptions. What are the five levels of AI autonomy? My autonomy ladder has five rungs: Macro Recorder, Copilot, Coworker, Agent and Autopilot Employee. Human involvement shifts from clicking and confirming to setting strategy and managing a fleet of AIs. Why stay model-agnostic? Models change. I want the business logic captured independently of the model, compiled into the instructions the active model needs, and re-tested whenever that model changes. --- ## The Dark Factory Canonical URL: https://www.brendanking.ca/dark-factory Author: Brendan King Last updated: 2026-11-01 The Dark Factory model for software development is AI-automated engineering in which people define and check the work while AI writes the code. Key points - The goal is a routine ticket built, reviewed, deployed and checked without hand-written code. - Today AI drafts bug fixes and finds causes; a person reviews the result. - Testing, a separate test environment and architectural records are foundations, not optional extras. The Dark Factory Routine bugs and features, handled completely by AI. The goal We want routine bugs and features to be handled completely by the Dark Factory. Written up as a ticket, built, reviewed, deployed and checked in production, with no hand-written code. What it is The Dark Factory is AI-automated engineering. Engineers define and check the work. AI writes the code. The name comes from manufacturing. A dark factory has no people inside, so the lights are off. But people do not leave the building. Their job changes. Why we are doing it "AI does the work" only means something if the work can move without us. If a person has to be involved for the work to happen, that person is the limit on how much we can build. The bottleneck keeps moving Same process, different constraint. First the limit was writing code. Then it was reviewing code. With the factory, the limit moves to what goes in: the quality of the brief, the prototype and the story. So people spend their time there. People work on the factory, not in it Three jobs: the quality of what goes in, tuning the factory, and judging what comes out. We review the combined output of the factory, not every single change, and use what we find to tune the machine. The best picture is a gardener. A gardener does not grow the plants. They tend the conditions. The AI does the work. The human tends the conditions it works in. The foundation is trust The factory only works on real foundations: a testing harness, a test environment that is not production, and a written record of architectural decisions. Review does not go away. It moves earlier, before the code, and later, after it is live. One process at a time Companies do not have an automation level. Processes do. Pick one process. Define what goes in, what comes out, and what good looks like. Automate it as far as it will go. Then take the next. You cannot automate a process that is not defined. Where we are today It is working on bugs now. AI drafts the fix and finds the cause, and a person reviews it. We are extending it to features and to more teams. The next step is a fix checked in production and the bug closed by the factory itself. Engineering after the factory The work moves earlier and gets harder. Get the inputs right. Then keep reshaping the systems, with the refactors that always lost to the next feature. The claim The Dark Factory makes software liquid. Liquid Software: https://www.brendanking.ca/liquid-software Questions about The Dark Factory What is the Dark Factory model for software development? It is AI-automated engineering. Engineers define and check the work; AI writes the code. The goal is to carry routine bugs and features from a ticket through production checks without hand-written code. Is the Dark Factory already fully autonomous? No. The current workflow described here has AI drafting bug fixes and finding causes, with a person reviewing them. Features, additional teams and automated production checks are the next steps, not claims of completed deployment. What do engineers do in a Dark Factory? People improve the inputs, tune the factory and judge its output. The bottleneck moves from writing code to the quality of the brief, prototype and story. What makes an automated engineering process trustworthy? The foundations are a testing harness, a test environment separate from production and a written record of architectural decisions. Review moves earlier, before code, and later, after it is live. How does the Dark Factory relate to Liquid Software? Liquid Software describes the business experience: software adapts to the job. The Dark Factory describes the engineering model intended to build and change that software. --- ## AI Principles Canonical URL: https://www.brendanking.ca/ai-principles Author: Brendan King Last updated: 2026-11-01 AI Principles is my set of ten operating principles for delegating work to AI while humans own direction, judgment and oversight. Key points - Assign outcomes, not questions; measure completed work rather than time saved. - Approval is adjustable: do it, do it and tell me, or ask first. - Stay model-agnostic, preserve memory and win through partnership. AI Principles The principles I run Vendasta on. 1. AI does the work. Humans orchestrate. Assign outcomes, not questions. 2. The AI leads and has an opinion. 3. Conversation is the front door. The software recedes behind it. 4. Approval is a dial, not a gate. Do it. Do it and tell me. Ask me first. 5. Memory is the job. A correction is remembered once and applied forever. 6. Routines are created by saying them. 7. Measure work completed, not time saved. 8. Use first principles to think 10X. 9. Maximize optionality. Minimize lock-in. We stay model-agnostic. 10. Win through partnership. Delegate, don't chat If the default button on an AI employee is Chat, it is a persona on a chatbot. A real AI employee takes a job and runs with it. Evolution of AI: https://www.brendanking.ca/evolution-of-ai Contact: https://www.brendanking.ca/contact Questions about AI Principles What is the central AI principle? AI does the work. Humans orchestrate. That means assigning outcomes instead of asking questions and keeping human attention on direction, judgment and oversight. What does “approval is a dial” mean? Different jobs need different levels of permission: do it, do it and tell me, or ask me first. Approval should reflect the job rather than blocking every action in the same way. How should AI work be measured? Measure work completed, not time saved. A real AI employee takes a job and runs with it; a chat interface alone does not establish that work was done. What is Vendasta’s 80/20 AI orchestration model? In my October 9, 2025 essay, I describe allocating 80 percent to AI execution and 20 percent to the oversight layer that directs it. It is a resource-allocation model, not a promise that every job is 80 percent automated.