You Set the Rules, The Work Gets Done
The start of a System of Agency for business automation
You know the follow-up that died because nobody owned it, the handoff that broke because the rule lives in one person's head, the customer exception that got dropped because the system had no place to put it. The work piles up in the gaps between tools, and after a while the job starts to feel like a trap — not because you lack judgment, but because your judgment has no clean path into how the work actually gets done.
System of Agency: You set the rules, and the work gets done.
— Rob Floyd, Owner, Eikon Digital Solutions
I know that feeling because I have lived it, and I know it shows up at every level of the business. Whether you are an owner/operator, a C-suite leader, a middle manager, or a person in the trenches, the weight is the same: you carry the meaning the system cannot hold, and it grinds the joy out of work that should matter to you.
When I set out to build what became BOSNet.io, the thing I wanted to fix was simpler than governance or AI architecture. I wanted people to enjoy running their businesses again. I wanted the person who sees the problem, knows the fix, and cares about the customer to have a real path into how the work gets done — instead of fighting the tools that are supposed to help. That is what turned this from an AI product into a System of Agency: software built to help people set rules, contribute judgment, and move work without forcing everything through one overloaded person.
The old software bargain
The old software bargain was simple enough: buy the app, set it up, keep it updated, and use it to run the business. That was better than paper, better than sticky notes, and better than trying to run customer relationships out of a personal inbox.
But those platforms were built around records, not relationships. The CRM holds the contact, the scheduler holds the appointment, the email platform holds the message, and someone still has to know what the customer actually meant, what was promised, and what needs to happen next. The software was never designed to carry that knowledge — it was designed to store files and let people do the rest.
The software holds the records. People hold the meaning.
The gap is not small. It is the part of the business that makes the work work, and most software has treated it like a side issue instead of the place where the real work starts.
Productiv put average managed SaaS spend at $9,600 per employee in 2023 and found that only 47% of licenses were used over a 90-day period. That is enterprise portfolio data, so it should not be treated like a perfect picture of every small business, but the same pattern shows up lower in the market too: smaller companies still carry too many tools once email, CRM, scheduling, payroll, marketing, payments, websites, social media, reporting, and industry-specific software all get counted.
The number is not the point by itself. The point is that tool count turns into memory loss, and memory loss turns into people inside the business becoming the place where the work has to be reconnected.
AI made the old bargain look new
AI made business software feel new for a while because the demos got better. The system could draft the email, summarize the meeting, rewrite the proposal, classify the lead, generate the post, and explain the dashboard without making the user click through as many screens.
I use AI every day, so I am not throwing rocks from outside the building. The problem is that adding AI to platforms built around records and workflows does not change who carries the work. It makes the old job faster, but the job stays the same.
Someone still reviews the draft, checks the summary, catches the missing context, decides whether the suggestion makes sense, and fixes the thing the tool missed because the business never had a proper place to put that knowledge. AI helps, but the work still runs through the same overloaded human checkpoint.
MIT's 2025 GenAI Divide report said 95% of enterprise GenAI pilots failed to produce measurable profit impact. I do not read that as proof that AI does not work; I read it as proof that putting AI on top of the same filing-cabinet way of working was never going to be enough.
Adding AI to the same old software makes the output faster, but the real work barely moves. Someone still configures, manages brittle handoffs, designs, and holds everything together by hand — AI just gave them a better first draft to fix.
If AI sits on top of a filing cabinet, it becomes a faster filing assistant. Maybe a very impressive one, but still a filing assistant.
Agency is the point
The shift that changed how I think about the platform is simple. AI as a Service is not wrong, but it is not the center of the idea. The stronger idea is a business that can set its rules, hear what people closest to the work are seeing, and have agents use those rules to carry real work for the people responsible for getting things done.
As the business changes, those should be rule changes, not code changes. The business should not need to rebuild software every time a new policy, customer exception, approval rule, service line, membership offer, or team responsibility changes.
I keep coming back to the phrase System of Agency because the point is not to remove people from the business; the point is to give people inside the business a stronger voice in how the work gets done, while moving them away from configuration work and repetitive execution in so-called automation tools.
A System of Agency means the business sets the rules, agents do the work, and people at every level can see, shape, and contribute to the work that moves the business forward.
The definition matters because AI acting without clear limits is not business automation. It is a raccoon with the office keys, and nobody should build a company around hoping the raccoon respects the budget.
Agency means the owner is not the only person who can move the business. It means the personal assistant who sees the missed follow-up, the department manager who knows the handoff is broken, the contractor who understands the customer expectation, the agency that owns a campaign, and the frontline staff member who hears the same complaint every week can all help shape the rules the agents work from.
People should not disappear into the system
Software built around records and workflows asks people to disappear into the tool. The person knows something useful, sees a problem, or has a better way to handle the work, but the system only gives them a form field, a status dropdown, or a comment box nobody reads. The platform was never built to hear from the people using it — it was built to process transactions and move data between screens.
It is a strange way to run a business. The people closest to the work usually know where the drag is, where the customer gets frustrated, where the handoff breaks, and where the process makes no sense.
The problem is not that those people lack ideas. The problem is that most systems do not give their ideas a clean path into how the business actually runs.
The people closest to the work should not be trapped at the edge of the system.
The Disciplines of Execution talks about people becoming active contributors instead of passive participants in the goals of the organization. I like that frame because the best business systems should not bury people under tasks; they should help people connect their judgment to the work that matters.
Agentic AI should be doing exactly this in a business. It should not just write faster emails or make prettier dashboards; it should help the business turn human judgment into rules, actions, follow-up, and proof.
What that means for real businesses
For a regional remediation services company, this starts with new customer contacts. The useful work is not "generate marketing content" as a vague activity; the useful work is finding potential customers, starting conversations, tracking the next step, and keeping the opportunity from disappearing when the team gets pulled into the day.
For a local winery, the work has a different shape. A beautiful place still needs steady attention around it through social posts, event promotion, customer signups, membership offers, follow-up, and reasons for people to come back with friends.
Neither business needs another pile of dashboards pretending to be progress. They need the work to move, and they need software that remembers what happened after the first conversation instead of pushing that memory back onto whoever happens to be closest to the problem.
The point is not more software activity. The point is work that keeps moving after the first conversation.
For a larger business, the same idea shows up differently. The pain may sit in departments, teams, handoffs, approval chains, agencies, contractors, or disconnected tools, but the problem is still the same: the people who understand the work do not always have a clean way to shape how the work gets done.
Doing tasks is not the same as carrying work
An automation tool can send an email when a form comes in, create a task when a deal moves forward, or post on Tuesday morning if someone already wrote the post and loaded the queue. That kind of automation has value, but it is fragile because every path is hard-wired in advance.
The rule changes, the customer does not fit the path, the connection between tools breaks, or the person who built the automation leaves. Then everyone gets nervous about touching the thing because nobody knows what else might snap when one branch moves. That is the cost of platforms that require someone to wire every step by hand — the wiring breaks every time the business changes.
Most automation asks someone to define the path in advance. A System of Agency starts with what the business knows, what the business allows, what the agent can do, when the agent must ask, and what record needs to be left behind.
Automation follows a path. Agency carries the work when the path changes.
No single person should be the place where every missing rule, every broken handoff, and every customer exception goes to be remembered. The way the business runs should carry more of that weight.
Trust, but verify
Blind trust is a terrible way to use AI. I am sensitive to AI flattery, and I do not want a system that tells me my idea is brilliant with the same confidence it uses when it makes up a source or writes a sentence that sounds like a wet LinkedIn brochure.
Encouragement is not proof. If an agent sends a message, the business should know what rule gave it permission; if it schedules a job, the team should know what capacity it checked; if it recommends a campaign change, the people responsible should see the reason and the evidence.
Trust is not the same as proof. Business AI needs proof.
Governance in the useful sense is not a committee, a binder, or compliance theater. It is the memory and proof that let a business hand off work without losing its mind.
The new bargain
The old bargain made people do the work: configuration, handoffs, first drafts, repetitive actions. The new bargain lets people tell the system what to do, and the work gets done.
The system is proactive from the start. As people interact with it, it learns the business — the rules, the exceptions, the way work actually moves. Drafts get closer to final. Follow-up stops dying. Handoffs stop breaking. As the system gets smarter, the work gets faster, until — depending on the stakes — full automation takes over and the human reviews the proof instead of doing the work.
For SMBs, the owner stops carrying the business by themselves. For larger organizations, people at every level get agency — and the insights, expertise, and ideas that old code-first platforms kept buried start shaping how the business actually runs.
The new bargain is not that AI replaces people. The new bargain is that people tell the system what the business needs, and the system gets smarter every time they do.
BOSNet.io is being built around that bargain. You set the rules; the work gets done.
Part 1 of the System of Agency Conversation.
Next: Your Software Is a Filing Cabinet.