Do Not Confuse The Task With The Attempt
A repair request can be assigned to the right person and still be in the wrong place operationally. That sounds like a small distinction until you try to automate it. A property manager...
Field notes on operational AI
Practical essays on building AI systems businesses can trust.
Short public essays split between applied workflow practice and the reusable primitives underneath AI systems of action.
Do Not Confuse The Task With The Attempt
A repair request can be assigned to the right person and still be in the wrong place operationally. That sounds like a small distinction until you try to automate it. A property manager...
Read latest essayPractice
Applied workflow notes
Applied field notes on making AI safer, more useful, and easier to trust inside real workflows.
Validate The AI Opportunity Before You Promise It
A workflow can look like a great AI pilot while the proof is still scattered across people's inboxes. That is the gap I think many SMB operators will run into as AI moves from demos into...
Do Not Let AI Match Its Way Past The Stop Sign
A compliance packet can look complete while the business is still one bad match away from a shipment problem. A nutraceutical packager is trying to release a finished lot. The...
Do Not Ask AI To Remember The Company
A contract renewal can look simple until the question lands in the middle of the real business. A commercial cleaning company is renewing a site contract. The signed agreement lists one...
The Record Can Exist Before The Work Is Ready
A new client folder can make a workflow look ready before it is safe to start. That happens in small businesses all the time. A professional services firm gets excited about a new...
Keep An Answer Key Under The Workflow
A workforce training provider can have a paper sign-in sheet on the front desk, a video attendance report from the remote class, a registration form in one folder, and a certificate...
Before AI Recommends The Next Step, Check The Source Health
A schedule can look open while the business is not actually allowed to use it. That is the kind of AI failure I think more operators are going to see as software starts recommending...
2026-08-01
Do Not Let AI Guess Which Records Are The Same Duplicate records are rarely clean duplicates. An insurance agency gets a quote request from a business owner through the website. The same...
Do Not Let Delegated AI Work Disappear Into The Chat
A medical billing appeal can be late before anyone notices the AI did anything wrong. The denial PDF is in a payer portal. The original claim submission proof is missing. The appeal...
Do Not Let The Plan Drift Quietly
Most small business plans do not fail in one dramatic moment. They drift. A supplier invoice gets paid twice because one copy went to AP and another went to the branch manager with a...
Your CRM Should Not Forget The Work Before The Customer Exists
A lot of customer work starts before the system has a clean place to put it. A prospect fills out a form. Someone else sends the budget in a call. A third person shares brand assets...
Do Not Let AI Import The Mess Faster
A batch upload can feel like an obvious place to use AI. The file is structured enough to parse, repetitive enough to be boring, and expensive enough to clean by hand. But the first...
Make AI Changes Easy To Inspect Before You Make Them Easy To Undo
A customer calls a machine shop and says the replacement parts were made to the wrong revision. Someone checks the drawing, looks at the purchase order, opens the complaint thread, and...
Know Who The Person Is Before AI Writes The Follow-Up
A customer follow-up can be wrong before the first sentence is written. Not because the tone is bad. Not because the grammar is off. Because the system does not know who the person is in...
Make The Artifact Carry The Proof
A customer-facing document can look professional and still create operational risk. That shows up in small businesses all the time. A quote is clean, but it was generated from the wrong...
Do Not Review From The Wrong Version Of The Work
A customer follow-up can look careless in the review queue even when the employee did the right thing. That sounds like an AI accuracy problem, but often it is a source problem. The...
Do Not Let AI Work From The Wrong Version Of The Business
A customer follow-up can be wrong even when every sentence is clear. Think about a veterinary clinic after a dental procedure. The pet goes home, a technician drafts discharge...
Make The Screen Show The Work
One thing I keep noticing in operational AI work is that the interface often needs to get denser after the product gets more useful. That sounds backwards if you are used to consumer...
Do Not Let The Dashboard Lie For You
One quiet way software creates operational risk is by making uncertain information look exact. You have probably seen this in a report, map, CRM view, invoice queue, or scheduling...
Map The Work Before You Automate It
The best AI opportunity in a small business often shows up as a person, not a process name. Ask an owner where the work hurts and you may hear "admissions," "follow-up," "billing,"...
Measure The Workflow, Not The Worker
A manager can watch an employee use AI to write a customer email in thirty seconds and still have no idea whether the business got better. The email may be clearer. It may be faster. It...
Make Approval Show Its Work
A lot of operational risk hides in the moment before someone clicks approve. The screen says the workflow is ready. The system has filled the form, prepared the message, matched the...
Know Who Is Allowed To Say Yes
A lot of small business workflows break in a place that looks too ordinary to call a systems problem: nobody is sure who is allowed to approve the next step. That can happen on the shop...
Stop Splitting The Review Work By Channel
A lot of small business follow-up gets messy because the work arrives through whatever channel the customer happened to use. Someone texts the owner photos from an install. Someone else...
Stop Treating Every Caller Like A Stranger
A lot of small business phone work starts with a tiny waste of trust. Someone calls back about the same estimate, invoice, sponsorship, repair, or delivery they discussed yesterday. The...
Make The System Update Reviewable
A small business can make the right decision and still lose the value of that decision. The customer approves the extra work. The manager agrees to the pricing change. The service...
Make The Phone Call Reviewable
A phone call feels complete when both people hang up. In a small business, that is often the dangerous moment. The customer thinks something changed. The employee thinks they captured...
Stop Asking Which Emails Are Important
Most small businesses do not have an inbox problem. They have an action-state problem hiding inside the inbox. A message from a customer can be urgent, but only if the order is blocked....
Start With The Workflow The Business Can Review
A small manufacturer can lose money in a surprisingly ordinary way: the owner approves a material surcharge by text, the team keeps production moving, and nobody turns that approval into...
Do Not Let One Example Become The Whole Workflow
One of the quiet ways AI projects go sideways is that the team keeps testing the same example. At first, that example is useful. It gives everyone something concrete to react to. A...
Start With The Intake Gate
Most businesses already have an invisible workflow engine. It is email. A customer sends a note. A vendor forwards a document. An employee replies from a thread that started three weeks...
Make The Change Visible Before You Automate It
One of the fastest ways to make an operator nervous is to show them an AI workflow that can change business records without showing exactly what changed. The fear is not irrational. In...
Let The Business Launch Empty
One thing I keep seeing in operational AI work is that the first trust problem often appears before the system has done anything useful. It appears during setup. The form asks for too...
Before You Add AI, Check The First Exception
Most workflow plans look cleaner on paper than they do inside the business. The intake process has an approval step, except one person knows which cases skip it. The billing process has...
Start With The Questions The Business Already Asks
Most operators I talk to do not start by asking for an autonomous AI agent. They start with a simpler frustration: they cannot get a clean answer to basic operating questions without...
Make The Rebuildable Part Obvious
One of the most practical questions an operator can ask before using AI in a real workflow is not, "Can the AI do this task?" It is, "If this task goes wrong, what can it touch?" That...
Start With Read-Only AI
The AI workflows that make operators most nervous often start in the places where the real business lives: inboxes, text threads, call logs, shared spreadsheets, and notes from someone...
Give AI A Clear Place To Work
Most operators do not start by asking whether an AI system can reason. They start with a more practical fear: if I give this thing access to the business, what exactly is it allowed to...
Start With What The Meeting Forgot
Most small-business meetings create work that never makes it into software. Someone explains the exception. Someone promises to follow up. Someone mentions that a customer always needs a...
Primitive
System primitives
Systems essays on the reusable product, architecture, and operating-model ideas behind the work.
Execution-Attempt State Is An AI Operations Primitive
Most business software treats work as a task with a status. That was good enough when humans were the invisible runtime. A person could look at a task, remember what they tried, know who...
Opportunity Validation Is The Bridge Between Finding And Commitment
The pattern I keep seeing is that AI systems are getting better at finding opportunities than proving which opportunities are ready to become commitments. That sounds like a...
Match Eligibility Is The Primitive Behind Safer AI Operations
A lot number can match and still not be enough. That is the part I think many AI workflow products will underestimate. Matching feels like a scoring problem. Find the certificate that...
Context Boundary Is The Missing Layer In Company-Aware AI
A company-aware AI assistant sounds simple until it has to answer a question the business would actually act on. "Can we renew this contract?" is not a pure language problem. The answer...
Provisioning State Is The Pre-Action Primitive
The pattern I keep seeing is that AI systems are better at creating records than knowing whether those records are ready for action. That sounds like a small implementation detail until...
Reference Truth Is An AI Operations Primitive
A transcript can sound complete and still leave the product guessing. Scans, screenshots, saved outputs, call notes, uploaded forms, and replayed conversations create the same risk. They...
Source-Health State Is An AI Operations Primitive
The next generation of AI operating systems will not be judged only by whether they can recommend the next action. They will be judged by whether the business can tell when the...
2026-08-01
Possible-Duplicate State Is An AI Operations Primitive The pattern I keep seeing is that business software treats sameness too bluntly. Two records are either duplicates or they are not....
The Task Ledger Is The Missing Primitive For Delegated AI Work
The more AI moves from answering questions to doing work, the less useful the conversation history becomes as the system of record. A conversation can start the work. It can explain the...
The Control Loop Is The Missing AI Planning Primitive
The pattern I keep seeing is that AI planning tools are too eager to become the source of truth. They summarize the work. They rank the tasks. They produce a daily plan. Sometimes that...
Lifecycle State Is An AI Operations Primitive
A customer account is often the wrong object for AI to reason from. The account tells the system who the business is dealing with. It does not reliably tell the system what can safely...
Intake Reconciliation Is An AI Operations Primitive
The pattern I keep seeing is that AI does not only make work easier to do. It makes messy inputs easier to ingest. That sounds like progress, and often it is. A model can read a CSV,...
Reversal State Is An AI Operations Primitive
The pattern I keep seeing in operational AI work is that the first hard question is not whether the system can change a record. It is what happens after it changes the wrong one. A...
Relationship-Role State Is An AI Operations Primitive
The pattern I keep seeing in operational AI work is that the contact record is usually too thin. Most business software can store a name, company, email address, phone number, title,...
Claim-Support State Is The Missing Layer In AI-Generated Artifacts
The more AI gets used to create business artifacts, the more I think the artifact itself is the wrong unit of trust. A deck, proposal, quote, customer update, invoice note, or follow-up...
Source Fidelity Is An AI Operations Primitive
The pattern I keep seeing in AI-assisted operations is that the system often reviews a projection of the work, not the work itself. That distinction sounds small until the projection...
Business Canon Is An AI Operations Primitive
The more AI systems touch recurring business workflows, the less I think "knowledge base" is the right frame. Most businesses do not only need a place to store documents. They need a way...
Operational Density Is An AI Interface Primitive
The pattern I keep seeing in AI workflow products is that the interface has to change once the system starts touching real operations. Early product screens often optimize for...
Uncertainty State Is An AI Operations Primitive
The pattern I keep seeing in operational software is that interfaces often prefer a clean answer over an honest one. A map wants a dot. A CRM wants a territory. An invoice system wants a...
Workflow Discovery State Comes Before Agents
One pattern I keep seeing in operational AI work is that the use case is rarely where the label says it is. "Renewal follow-up" sounds like messaging. "Warranty follow-up" sounds like...
Workflow Outcome State Is The AI Measurement Primitive
The mistake in many AI ROI conversations is that they inherit the old productivity unit: the individual worker. That unit was already weak in knowledge work. Lines of code, emails sent,...
Approval Evidence Is An AI Operations Primitive
The more AI systems move from drafting into acting, the less satisfied I am with the phrase "human in the loop." It sounds safe, but it hides the important product question: what exactly...
Authority State Is The First AI Operations Layer
The more operational AI work I see, the more I think identity is being underdefined. Most teams treat identity as login. A user signs in. A session exists. Maybe there is a role. Maybe...
Evidence-Specific Review Is An AI Operations Primitive
One pattern I keep seeing in operational AI work is that the channel is usually the wrong abstraction. A customer calls. Another texts. A PDF arrives. A voicemail contains the missing...
Identity State Is The Missing Layer In AI Phone Workflows
Phone workflows make one limitation of current AI products obvious: recognizing a person is not the same thing as knowing what the system is allowed to do with that recognition. A number...
Governed Projection Is The Missing AI Operations Layer
A lot of operational AI work looks finished too early. The assistant reads the message. It extracts the decision. It summarizes the call. It creates the task. It routes the handoff. From...
The Call Is Not The Product
Most voice AI demos are evaluated at the wrong layer. The obvious question is whether the agent sounds natural. Does it interrupt politely? Does it recover from confusion? Does it avoid...
Attention State Is An AI Operating Primitive
The pattern I keep seeing in operational AI work is that the hard part is rarely "read this message." The hard part is knowing what state the work is in after the message is read. Email...
Practice And Primitive Are Different AI Products
A text message approving a material surcharge does not look like product architecture. It looks like a small business keeping work moving. The owner replies yes. The shop orders the...
Scenario Inventory Is Becoming An AI Operating Primitive
The part of AI workflow design I keep paying more attention to is the example set. Not the polished demo examples. The real ones. The duplicate request that arrived through two channels....
Deterministic Intake Is A System-Of-Action Primitive
The pattern I keep seeing in operational AI work is that the messy part starts before the model ever gets involved. Work arrives through email threads, forwarded documents, customer...
Audit Is Becoming A System-Of-Action Primitive
The pattern I keep seeing in operational AI work is that the hard part begins after the model produces a good answer. A recommendation is easy to admire in a demo. A state change is...
Onboarding Is Becoming An Operating Primitive
The part of AI products I am starting to pay more attention to is not the demo. It is setup. Setup looks mundane. Create an account. Invite users. Choose a plan. Connect a few...
Plan Linting Is An AI Operating Primitive
The pattern I keep seeing in real AI implementation work is that the plan often fails before the model does. The architecture may be directionally right. The product idea may be useful....
Operational Intelligence Is The Vertical AI Wedge
The pattern I keep seeing in vertical software is that buyers may ask about AI, but they lean in when the conversation gets to operational visibility. Can the system tell us what is...
Disposable Runtime Is An AI Operating Primitive
The pattern I keep seeing in real AI implementation work is that "permission" is too small a word for what agents actually need. Most product discussions treat permission as access to...
Permissioned Retrieval Is The First AI Operating Primitive
The pattern I keep seeing in real AI automation work is that the interesting data is rarely sitting in a clean product table waiting for a model. It is in inboxes, messages, call logs,...
Scoped Context Is The Operating Primitive
A lot of AI product discussions still start with the model or the interface. Can the model reason? Is the chat experience good? Can the assistant retrieve the right documents? Those...
The Work Starts Before The Software Sees It
Most business work starts before software can see it. It starts in a meeting, a call, an exception someone explains out loud, a follow-up everyone assumes someone else wrote down, or a...