The Invisible Factory

The Problem Nobody Talks About

In 1950, Toyota had a problem that every manufacturer shared but few acknowledged: waste was invisible. Inventory piled up between stations. Defects flowed downstream undetected. Workers waited for parts with no way to signal it. The factory floor was a black box, and management made decisions based on reports that were already outdated by the time they arrived.

Taiichi Ohno’s genius was not that he invented better machines. It was that he made the invisible visible. Andon cords let any worker signal a problem in real time. Kanban boards made workflow visible at a glance. Standardized work documented the best known method for every task. Once you could see the waste, you could eliminate it. Once you could see the flow, you could optimize it.

Seventy-five years later, knowledge work has the exact same problem — but worse.

The Black Box of Knowledge Work

In a factory, you can at least walk the floor and see inventory stacking up. In knowledge work, you literally cannot see what anyone produces unless they show you. Work lives in private AI conversations, personal folders, scattered documents, and messaging threads that vanish into scroll. There is no factory floor to walk.

A CEO managing ten people has no way to answer the most basic operational questions:

• What did each person actually produce today? Not what they said in standup — what did they build?
• Is that work aligned with our quarterly goals? Or are people drifting toward interesting but irrelevant tasks?
• Is the quality improving? Or are we making the same mistakes we made last month?
• What’s blocking progress? And can we catch it at 9am instead of discovering it at 5pm?
• What actually shipped? Not what’s in progress — what created value?

These are not sophisticated questions. In manufacturing, they have been answered in real time since the 1960s. In knowledge work, most companies answer them once a quarter — if at all.

The AI Paradox

The arrival of AI should have made this easier. Instead, it made the problem worse.

AI gives every employee access to extraordinary capabilities. A single person with the right prompts can produce work that used to require a small team. But that work happens inside private AI conversations. The prompts are ephemeral. The outputs scatter across personal folders. The institutional knowledge that was generated — the strategies, the templates, the decision logic — evaporates the moment the chat window closes.

Companies are spending thousands of dollars per month on AI tokens, and they have no way to tell if the investment is working. They can see the bill. They cannot see the output.

What Toyota Would Do

If Taiichi Ohno walked into a modern knowledge work company, he would see exactly what he saw at Toyota in 1950: invisible waste everywhere. People duplicating work because they can’t see what others have already built. Quality drifting because there are no shared standards. Decisions delayed because executives are working from status reports instead of actual output. Improvements that never compound because nobody captures what worked.

He would also see the solution clearly: make the work visible. Build a system that captures real output, compares it against goals, and creates a daily rhythm of small improvements that compound over time.

That system is GAAPSS.

Six Folders That Organize Any Role

The Grammar

GAAPSS is a six-folder structure that organizes any role in any department. Every employee, from the CEO to the newest hire, maintains the same six folders. The structure is simple enough to set up in five minutes and powerful enough for AI to review every night.

The acronym spells GAAPSS — pronounced like “gaps,” which is fitting, because the framework exists to close the gaps between strategy and execution, between activity and output, between what you planned and what actually happened.

Goals: The Compass

Every employee’s Goals folder contains their quarterly rocks, key performance indicators, and success criteria. These are not daily deliverables — they are static reference documents that define what success looks like this quarter.

The Goals folder answers one question: What should this person be working toward? Without it, daily work has no anchor. With it, every overnight review can measure alignment: is today’s work moving the needle on what matters, or is it drift?

Goals are updated quarterly when rocks change. There is no need for daily reports here — the AI review agent generates alignment insights by reading Goals alongside Projects and Shipped.

Advisors: The Expertise

Advisors are encoded knowledge files — markdown or HTML documents that contain expert context, scoring rubrics, and domain-specific best practices. Think of them as the best practices of a role, made machine-readable.

A great advisor has three components:

1. Expert context — who the advisor represents and what they know. A landing page optimization advisor might encode the principles of a world-class conversion rate optimizer.
2. A scoring rubric — how to evaluate work against standards. Specific criteria, weighted by importance, that an AI can use to assess quality consistently.
3. Actionable guidance — concrete principles, not vague platitudes. “Headlines should be under 8 words with a clear benefit” beats “Write good headlines.”

Advisors serve two purposes. First, they help employees produce better work by giving them access to encoded expertise while they work. Second, they give the overnight review agent a standard to measure against. When Humans+Agents reviews a landing page and finds it doesn’t match the optimization advisor’s rubric, that becomes a coaching opportunity — not a penalty, but a specific suggestion for improvement.

Agents: The Workforce

Agents are repeatable workflows that have been standardized enough to delegate to AI. Each agent lives in its own subfolder and contains at minimum two files: a process definition (how to do the task) and a golden example (what excellent output looks like).

The golden example is the secret ingredient. LLMs are extraordinarily good at matching a reference. When you show an AI “here is what a great client proposal looks like” alongside “here is how to make one,” the output quality jumps dramatically compared to instructions alone.

You should create an agent when you’ve done a task more than three times and the quality matters. If you find yourself explaining the same process to a colleague or to ChatGPT, it’s ready to become an agent.

Projects: The Work

Projects is where daily work lives. Each active project gets its own subfolder containing the real artifacts being built — HTML files, documents, images, data, prototypes, deliverables. This is the folder that changes every day, and it’s the primary input for the overnight review.

Humans+Agents diffs the Projects folder daily. It tracks new files, modified files, and the nature of the changes. It compares what was produced against Goals (alignment), Advisors (quality), and Agents (standardization). The overnight review tells the CEO: here is what each person worked on, here is how it aligns with their goals, and here is where coaching could help.

Signals: The Intelligence

Signals is the intelligence layer — everything that provides context for better decisions. Market observations, competitive intelligence, customer feedback patterns, internal insights, blockers, opportunities. This is also where context lives: background information that helps the AI review agent understand the “why” behind the work.

Signals feed the overnight review with information that Goals alone cannot provide. When Humans+Agents sees a project that seems misaligned with stated goals, it checks Signals for context — maybe there’s a strategic reason. When the CEO sees an unexpected pattern in the morning briefing, Signals provide the explanation.

Shipped: The Output — and the Learning

Shipped is the most important folder in the framework, because it is the one that separates activity from output — and captures the knowledge that makes tomorrow better than today.

Work moves to Shipped when it creates value: deployed to production, delivered to a client, handed off to the next person in the chain. The distinction between Projects (what you’re working on) and Shipped (what created value) is what makes the framework’s most important metric — ship rate — possible.

But Shipped is not just for finished deliverables. It is also for learnings.

Every day, as you productively spend tokens and push work forward, you encounter things worth capturing: a bottleneck you overcame, a challenge you didn’t expect, a technique that worked better than the old way, a tool configuration that saved an hour, a prompt pattern that unlocked better output. These learnings — big and small — belong in the Shipped folder alongside your finished work.

Moving finished work to Shipped is a conscious act. It forces the question: did I actually deliver something, or did I just stay busy? Many things get shipped but never moved here — that’s a discipline problem, not a shipping problem. Building the habit is part of the daily cadence.

Shipped accounts for nuance. Internal handoffs count. Updates to existing production assets count. A learning document with three sentences about a bottleneck you solved counts. The goal is not to ship everything — it’s to make the ratio visible and the knowledge transferable so the whole team can improve.

The Daily Cadence

The 24-Hour Loop

GAAPSS is not a filing system. It is a production system. The folder structure only matters because of what it enables: a closed-loop daily cadence that compounds improvements every 24 hours.

The cadence has six phases, and it runs every single day, forever.

Phase 1: Publish (End of Day — 2 minutes)

At the end of every workday, each team member publishes their GAAPSS folder. In the simplest version, this means dragging the folder into Humans+Agents. The system captures what changed since yesterday — new files, updated projects, items moved to Shipped.

This is the “Publish Today” ritual. It takes two minutes. It is the single most important habit in the system, because without it, nothing else works. No visibility, no review, no improvement.

Phase 2: Overnight AI Review (Automatic)

While the team sleeps, the CEO’s review agent reads every published folder across the entire company. For each employee, it:

• Reads all published folders to understand what was created, updated, or shipped that day.
• Compares work against Goals to measure alignment. Is this person’s daily output moving them toward their quarterly rocks?
• Checks Advisor standards to evaluate quality. Does this landing page meet the optimization advisor’s scoring rubric?
• Surfaces patterns and blockers to identify coaching opportunities. Are two people duplicating work? Is someone stuck on the same problem for the third day?
• Measures ship rate to track the ratio of Projects to Shipped across individuals, teams, and departments.

The output is a morning briefing for the CEO: a prioritized summary of what happened, what needs attention, and where the improvement opportunities are.

Phase 3: Stand-Up (First 15 Minutes)

The morning starts with a 15-minute operations review. This is not a traditional stand-up where people recite what they did yesterday — the AI review already knows what happened. Instead, the CEO shares the overnight analysis with the team. Findings. Goal alignment. Quality observations. Patterns.

Critically, the stand-up also includes learnings from yesterday’s Shipped folders. When someone overcame a bottleneck, discovered a faster workflow, or figured out how to get better output from an agent — the whole team hears about it at 9am. This is the social learning cadence in action: the organization literally gets smarter every morning. Yesterday’s hard-won insight becomes today’s starting advantage for everyone.

Everyone leaves with a clear goal for the day — not a vague intention, but a specific, measurable objective informed by the overnight review — and new knowledge from their teammates that they did not have the day before.

Phase 4: Execute (First Hour)

Significant progress is expected within the first hour. This sounds aggressive, but it is the natural consequence of two things: clear direction from the stand-up, and the compressed capability that AI provides. A person who knows exactly what to build, with advisors and agents at their disposal, can produce in one hour what used to take a day.

Phase 5: Coach (Second Hour — 15 minutes per person)

If significant progress has not happened within the first hour, or if someone hits a blocker, the CEO books a 15-minute coaching session in the second hour. This is the critical innovation: help arrives at the start of the day, not the end.

In most organizations, a blocker discovered at 9am is not addressed until the next day’s stand-up — or worse, the next week’s meeting. In the GAAPSS cadence, coaching happens by 10am. The rest of the day is productive instead of wasted.

Phase 6: Deliver (Rest of Day)

The team executes with clarity for the rest of the day, producing real artifacts inside their GAAPSS structure. At the end of the day, they publish again. The loop closes. The cycle continues.

Why Daily Matters

Most operating frameworks run on a weekly or monthly cycle. EOS has weekly L10 meetings. OKRs are reviewed quarterly. Scrum runs in two-week sprints. GAAPSS runs daily.

The difference is not incremental — it is exponential. A weekly feedback loop means you catch a problem on day 7 and fix it on day 8. A daily feedback loop catches it on day 1 and fixes it on day 2. Over a quarter, that’s not 4x faster — it’s fundamentally different, because early problems don’t cascade into larger ones.

Consider compound improvement. If you improve 1% per week, you’re 1.52x better after a year. If you improve 1% per day, you’re 37x better. The math is the same math that makes compound interest powerful, and it applies just as forcefully to operational excellence.

Ship Rate

The Metric That Matters

Of all the metrics GAAPSS makes possible, one stands above the rest: ship rate. It is the ratio of work started to value delivered — the percentage of your Projects folder that makes it to Shipped.

Ship rate answers the question every CEO should be asking but rarely can: of all the work my team is doing, how much of it is actually creating value?

Why Ship Rate Works

Most productivity metrics measure inputs: hours worked, tasks completed, meetings attended. Ship rate measures output. It does not care how many hours something took or how many tasks were checked off. It asks one question: did it ship?

This matters because knowledge work has a unique failure mode: people can be extremely busy while producing very little value. A team can have 37 active projects and only 8 shipped deliverables — a 22% ship rate. That means 78% of the team’s energy is bound up in work-in-progress that has not yet created value.

In manufacturing, this is called inventory — and Toyota identified it as one of the seven deadly wastes. In knowledge work, it has never had a name. GAAPSS gives it one: the gap between Projects and Shipped.

Measuring Ship Rate

Because GAAPSS separates Projects and Shipped architecturally, ship rate is measured automatically. Humans+Agents counts the items in each folder and calculates the ratio. The metric is tracked daily across three levels:

• Individual: Each employee’s personal ship rate. Are they finishing what they start?
• Team: The department’s aggregate ship rate. Is this group delivering?
• Company: The organization’s overall ship rate. Is the business converting activity into value?

A healthy ship rate depends on context. A creative team running many experiments might have a lower rate than an operations team with defined deliverables. The point is not to hit a specific number — it’s to make the number visible and improve it over time.

Ship Rate and the Daily Cadence

Ship rate improves as a natural consequence of the daily cadence. When blockers are caught at 9am instead of next week, fewer projects stall. When coaching happens in the second hour, people get unstuck before wasting a day. When the overnight review identifies duplicate efforts, resources are consolidated.

Over time, the daily cadence drives ship rate up — not through pressure, but through visibility and support. The system is designed so that helping people ship is the default behavior, not an exception.

The New Toyota Production System

A System, Not a Tool

The Toyota Production System did not succeed because Toyota had better machines. It succeeded because Toyota had a better system — a set of principles, practices, and daily disciplines that made waste visible and improvement continuous.

GAAPSS is built on the same insight. It is not a tool. It is a production system for knowledge work. And the mapping between TPS principles and GAAPSS practices is not a loose analogy — it is structurally precise.

Where GAAPSS Goes Beyond TPS

The parallel with TPS is structural and real, but GAAPSS goes beyond it in one critical dimension: capability is elastic.

In manufacturing, a machine’s capability is fixed. A CNC machine cuts at the speed it cuts. You can reduce waste around it, but the machine does not get smarter. Toyota’s system optimized a fixed-capacity environment.

In knowledge work with AI, every person’s capability is elastic. It is a function of how well they use tokens — the AI compute that powers their daily work. A person who learns to use tokens effectively does not just fix today’s problem. They permanently raise what they can produce tomorrow.

This means GAAPSS is not just eliminating waste in a fixed-capacity system. It is compounding capability in an expanding-capacity system. Every coaching session that teaches someone to use AI better permanently raises the productivity ceiling. TPS compounded efficiency. GAAPSS compounds capability.

The Copying Problem

TPS took Toyota decades to build and was famously difficult for competitors to replicate. Companies would visit Toyota, see the kanban boards, go home and implement kanban boards, and nothing would change — because they copied the artifacts without the discipline.

GAAPSS has a structural advantage here. The discipline is built into the system. You publish your folder or you don’t. The AI reviews it or it doesn’t. The stand-up happens with real data or it doesn’t. You cannot fake the daily cadence the way you can fake a kanban board. The folder is either there with real work in it, or it is empty.

This makes GAAPSS more transferable than TPS ever was. The system enforces its own adoption because the value is immediately visible: publish today, get a review tomorrow morning. The feedback loop is 24 hours, not 24 months.

The CEO as Operator

A New Mental Model

The traditional CEO mental model is: I manage people, and people produce output. The leverage question is always headcount — hire more, manage better, hope for productivity gains. But the cost of adding another salary is enormous, and the marginal productivity of the eleventh person is never what the third person was.

GAAPSS introduces a different model. Your most expensive line item is salary, but the thing that multiplies what each salaried person can produce is tokens — the AI compute they consume daily. A person spending $200 per month in tokens who knows how to use them can produce what used to require a small team.

The CEO’s job, then, is not to manage people in the traditional sense. It is to maximize the value creation rate per dollar of salary, and the lever for that is token utilization on productive work.

The $1,000-Per-Day Standard

In early 2025, the idea that employees should use AI or face consequences was controversial. By late 2025, it was becoming obvious. In February 2026, StrongDM’s AI team published what they learned building a “Software Factory” — a fully non-interactive development environment where specs and scenarios drive agents that write code, run harnesses, and converge without human review. Their practical benchmark:

One thousand dollars per engineer per day. That is not a theoretical ceiling — it is a floor that a real engineering team is hitting and calling insufficient. The economics of what a single person can produce have shifted that dramatically, that fast.

Now consider our context. Every employee in this company is on a Max plan — $200 per month. That felt like a significant investment when we started. Today, it is the absolute baseline. And here is the standard we are setting internally:

This is not about spending tokens for the sake of spending them. It is about productive utilization — tokens spent on work that appears in your Projects folder, that gets reviewed overnight, that moves to Shipped. The GAAPSS cadence makes this visible: are you leveraging AI to produce more, or are you doing things manually that a model should be doing instead?

StrongDM’s team distilled this into a single question every person should ask themselves throughout the day: Why am I doing this? The model should be doing this instead. That mindset — combined with the GAAPSS structure to capture and review the output — is how a team of ten produces what used to require fifty.

The CEO Review Agent

Because every employee publishes their GAAPSS folder daily, the CEO can build a personal review agent that reads the entire company every night. Not summaries. Not status reports. The actual artifacts.

The review agent follows a five-step process:

1. Read all published folders. Ingest every team member’s GAAPSS folder, understanding what was created, updated, or shipped that day.
2. Compare against goals. Measure each person’s work against their stated quarterly rocks. Flag drift. Confirm alignment.
3. Check advisor standards. Evaluate work quality against the scoring rubrics in each person’s Advisors folder. Enforce standards consistently.
4. Surface improvement opportunities. Compile patterns, duplications, blockers, and coaching moments into a morning briefing.
5. Measure ship rate. Track the ratio of Projects to Shipped across individuals, teams, and departments.

The result: the CEO wakes up every morning to a clear picture of what the entire company produced yesterday, how it aligns with strategy, where the coaching opportunities are, and who needs help today. This is genchi genbutsu — “go and see” — at a scale and frequency that no human manager could achieve alone.

Coaching, Not Surveillance

There is a critical distinction between using GAAPSS for coaching and using it for surveillance. The distinction is not philosophical — it is structural.

Surveillance systems monitor behavior: keystrokes, screen time, mouse movements, application usage. They answer the question “Is this person working?”

GAAPSS reviews output: artifacts, deliverables, shipped work. It answers the question “Is this person producing value, and how can we help them produce more?”

The difference matters because it changes the emotional experience of the system. A surveillance system creates anxiety. A coaching system creates support. When an employee knows that their manager will see their work tomorrow morning and offer specific, useful feedback by 10am, the system feels like having a great boss — not like being watched.

For EOS Companies

Closing the Gap Between L10s

If your company runs on the Entrepreneurial Operating System (EOS), GAAPSS is not a replacement — it is an accelerator. EOS provides a powerful weekly rhythm: the Level 10 Meeting, the Scorecard, Rocks, the Accountability Chart, and the Issues List with IDS (Identify, Discuss, Solve). GAAPSS takes that same structure and makes it daily.

The gap in EOS has always been the six days between L10 meetings. A lot can go wrong — or right — in six days, and the weekly meeting is the first time anyone finds out. GAAPSS closes that gap by making the review continuous.

EOS + GAAPSS in Practice

The weekly L10 does not go away. Instead, it becomes a strategic session that reviews weekly patterns rather than catching up on what happened. The daily stand-up handles operational awareness. The overnight review handles data collection and analysis. The L10 is freed to focus on the higher-order decisions that actually need 90 minutes of collective attention.

For EOS Implementers, GAAPSS provides something that has always been missing: a way to verify that the system is actually running between sessions. When a company’s GAAPSS folders are active — Goals populated, Advisors built, ship rate trending up — the EOS framework is working. When they’re empty, it’s not. The folders are an honest mirror.

Getting Started

Day One: Set Up the Folders

Create your GAAPSS structure. Six folders, numbered for sort order:

1. Goals — Add your quarterly rocks and KPIs.
2. Advisors — Start with one advisor for your primary domain.
3. Agents — Document one repeatable process you do frequently.
4. Projects — Move your active work here, one subfolder per project.
5. Signals — Add one file with current market context or blockers.
6. Shipped — Move anything that’s already in production here.

This takes five minutes. Do not overthink it. The folder structure will improve over time as the daily cadence surfaces what’s missing.

Week One: Establish the Publish Habit

The only objective for week one is daily publishing. At the end of every day, publish your GAAPSS folder. Do not worry about the overnight review, coaching sessions, or ship rate. Just publish.

If you are deploying GAAPSS across a team, make publishing the only ask for the first week. Do not introduce the full cadence yet. The habit of publishing daily is the foundation that everything else builds on.

Week Two: Start the Morning Stand-Up

Once publishing is habitual, introduce the 15-minute morning stand-up. Review what the overnight analysis found. Assign clear daily goals. Keep it to 15 minutes — the data is already there, so you are reviewing, not reporting.

Week Three: Add Coaching

Introduce the second-hour coaching sessions. Anyone who did not make expected progress in the first hour gets a 15-minute session. The key is speed: help arrives at 10am, not at the next weekly meeting.

Week Four: Measure

By week four, the full cadence is running. Now measure. Look at ship rate across individuals and departments. Look at goal alignment trends. Look at which advisors are being used and which are not. The data from four weeks of daily cadence will show you more about your team’s actual output than most companies learn in a quarter.

Making the Unseen Seen

The raw material of continuous improvement is visibility. You cannot fix what you cannot see. You cannot coach what you cannot read. You cannot compound what you cannot measure.

For seventy-five years, manufacturing has had this visibility. The Toyota Production System and its descendants gave factory managers the ability to see workflow, detect waste, and improve daily. Knowledge work never had an equivalent — until now.

GAAPSS makes knowledge work visible. Six folders, published daily, reviewed by AI overnight, discussed in a 15-minute stand-up, coached in real time. The structure is simple. The discipline is demanding. The results compound.

The question is not whether this is possible. The tools exist. The framework is here. The question is whether you will build the habit.

Publish today. Review tomorrow. Improve forever.