We are in an experimentation phase with AI. Expectations are sky-high, and as many people have observed, we may well be in a bubble.
Recent reports paint a sobering picture. RAND Corporation found that more than 80% of enterprise AI initiatives fail to deliver their intended business value, roughly double the failure rate of comparable non-AI IT projects. A separate MIT study found that 95% of generative AI pilots produce no measurable profit-and-loss impact. Gartner’s own survey of infrastructure and operations leaders found that only 28% of AI use cases fully succeed and meet ROI expectations, while 20% fail outright.
The reason most of these initiatives fail, in my opinion, is that organizations approach them technology-first. They buy subscriptions from a vendor like OpenAI or Anthropic, hand the tools to employees, and expect them to figure out what to do with them.
Employees learn to prompt the AI. They find it useful for drafting emails or analyzing data. But they never move beyond the chatbot.
SAFi takes the opposite stance. SAFi forces you to think about your processes before you can use it. At a minimum, you need a company charter, meaning a mission statement and core values, or a policy, before you can build an agent. Every SAFi agent needs a purpose, or, as we would say in philosophy, a teleology. You will know when you have not defined the agent’s teleology clearly, because the agent will be misaligned.
AGI, or Artificial General Intelligence, and the idea of AI agents being fully autonomous without people defining workflows and processes, is an illusion. You can tell the executives who have bought into the fantasy of autonomous AI because they start throwing terms around like “we are an AI-first company,” yada yada, when they themselves have not even used a chatbot.
That statement only makes sense when you are selling to other AI agents or machines. But when your audience and customers are people, human connection has to come first. A more accurate statement would be: we are a human organization empowered by AI.
When using SAFi, you need to step back and think about your workflows and the tasks you want to automate. You know, the things you wish you had an assistant for: daily reports, onboarding and offboarding processes, or any process you currently have. Once you have a mental picture of that process, or even better, have written it down, it becomes much easier to build an agent to automate most, if not all, of it.
You define the tools the agent will use and the data it will have access to, including read and write permissions. You do not simply let it run and expect the same result every time.
You always need to be in control, and SAFi gives you tools to do that. As a human, you possess abilities that nobody else can replace. Your messiness is not a defect. It is a feature. The routine and tedious work, however, is something you can offload to an agent.
Top comments (0)