Jacob SagerShow me your workflow
Architecture & Stewardship Standard · v2.0

How I design and look after a business system.

This is the standard I use when building software for a business. It covers where the records live, which tools connect, what needs human approval and how to check that a job finished correctly. I call the connected setup a Business Nervous System.

Architectural principles

The rules that shape a deployment.

Authoritative ownership

For each important kind of information, name the place people should trust. An invoice belongs in accounting, for example. Any copy elsewhere needs a purpose and a way to stay current.

Capture near creation

Capture information as close as practical to the event that creates it. Bad source data needs attention at the source.

Visible human judgment

Name the decisions that require a person, such as approving a customer commitment. Any decision software may make needs an agreed rule and checks.

Observable automation

If an automation fails, someone responsible needs to be able to see it. Agree where the error appears and what they should do next.

Named responsibility

A named person or role is responsible for the result of each important process.

Explicit completion

Define what finished means and how to check it. Creating a draft invoice and sending an approved invoice are different steps.

Deliberate integration

Keep tools that do their job well. Connect them where a process needs information to move between them.

Portability

Keep the documentation and account ownership clear enough for someone else to take over the system responsibly.

Reference architecture

Six connected parts of the system.

Six connected parts of a business system Company records connect to software that moves and prepares information, screens people can use, and checks that bring problems to someone responsible. The parts exchange information in several directions. Systems of record Accounts · customers · jobs Shared business records Shared records · current job status Integration Connections between software Processing Rules · calculations · selective AI Interface Task lists · screens · people Control Checks · errors · calls for help
Systems of record The software that knows its part.
Business memory The records and current status of each job.
Integration Information moves between systems.
Processing Details are prepared for the next task.
Interface Each person can see and do their part.
Control People can find failures and unfinished jobs.
The tools exchange information in several directions. Separate checks look for errors and unfinished jobs, then bring them to the person responsible.
Architecture × function

Which parts of the system do each job?

The functions below mean noticing incoming work (Sensing), sharing the needed information (Signaling), preparing it (Processing), assigning a decision (Deciding) and checking the result (Regulating). “Primary” marks the part that does most of that job; “Supporting” marks a part that helps.

How the system notices, shares and prepares information, supports decisions and checks results
Layer Sensing Signaling Processing Deciding Regulating
Systems of record Primary Supporting — — Evidence
Business memory Supporting Primary Supporting Context Supporting
Integration Event capture Primary Reformat Send to the right place Return status
Processing Sort Add details Primary Rules Check the result
Interface Observe Deliver Explain Human action Confirm
Control Monitor Monitor Monitor Ask a person Primary

The Notion standard

Notion is an app for shared documents and linked records. Where it fits, I use it to organize customers, jobs and instructions. Relationships connect the records; views let each person see the tasks and information relevant to them.

A template should represent a task that really repeats. Completed work can be archived without losing its history. If the workspace copies information from another tool, the design explains why and how that copy stays current.

The automation standard

Each live automation needs a name and a job. Document what starts it, which records it can change, who is responsible and what happens if it fails. Use accounts controlled by the client or clearly document who owns the service account.

Set a limit on automatic retries. Where a repeated request could create two invoices or duplicate messages, add a check that prevents the action happening twice. Changes to a field or source record need review if another process relies on them.

For an AI step, define what a usable answer looks like and what happens when information is missing or the result is uncertain. Decisions with important consequences need a route to a person.

Represent the operating state.

A job might move through these statuses: created, assigned, in progress, waiting, completed and checked. The business may need different words. What matters is being able to see what is happening and who needs to act.

Control checks the business outcome.

Look for jobs that have sat too long, tasks with nobody responsible and connections that failed. Compare records in tools that should agree, such as completed jobs and invoices.

A list of items needing attention can be more useful than a large dashboard. Someone should be able to see the problem and the next action.

Stewardship operating standard

The continuing responsibility.

Routine monitoring
Check whether automations ran and review reported problems on the schedule we agree.
Maintenance
Repair agreed connections, broken links and defects in the existing processes.
Change control
Record important system changes and which other processes they affect.
Small improvements
Make agreed changes to existing screens, record fields, rules and processes.
Periodic recalibration
Check whether the records, job statuses and approval rules still match how the business runs.
Documentation
Keep the system map and list of automations current, including the tools each one depends on.
Security and access
Give people and software the access their jobs need. Review it when responsibilities change, and avoid unnecessary copies of sensitive information.
Offboarding
Provide a current system map, list of automations and account-ownership notes so the next person can take over.

The engagement names the implementation, maintenance and continuing decisions included. Substantial new work requires an agreed scope, cost and owner before it begins.

AI operating standard

Agree what the system should achieve and what it may do.

Purpose and economics

Name the business result to test. Measure the current process before changing it, or explain why that measurement is unavailable. Keep build and running costs visible.

Data and company context

Identify the information the system may use and where to find the current version. Specify what to do when details are missing or out of date. Review supplier terms for the intended use.

Human decisions

Name the actions software may take, what a person must approve and who handles unusual cases. Giving the system more permission requires an explicit decision informed by testing.

Testing and continuing review

Test ordinary jobs and the unusual cases that could matter. Record failures and corrections. Keep a way to do the work if the system stops, a person to contact and criteria for revising or stopping the automation.

Keep a list of live systems with their purpose, responsible person, data sources, permissions and review schedule. Include instructions for operating them and responding to problems, so a suitable person can continue the work.

Definition of healthy

Important work reaches a verified outcome.

People can see new work, find the details needed to handle it and check what happened. A named person can find a failure or an unfinished job and do something about it.

Explore Stewardship →