Premium Access

The Modular Primitive

Brighton MlamboApril 28, 2026
Engineering the runtime. Why the future of work is a lego set of raw primitives.
Premium Access Only — $60.00 USD

"Our approach to AI agents is very lego-like. In a sense, we want to democratize and give modular access to the raw primitives of computing to our users and clients." ~ Brighton Mlambo

Access Locked

"Knowledge is the only asset that grows when shared, but strategy is only for those who protect it."

Preface: The Ghost in the Ledger

In 1887, Alexander Hamilton Church began obsessing over the cost of a spindle. He was a pioneer of Management Accounting, and he realized that the traditional way of counting money (Financial Accounting) was useless for actually running a business. Financial accounting told you if you were profitable at the end of the year; Management Accounting told you why you were losing money on every yard of fabric on Tuesday afternoon.

Church invented the concept of "Overhead Allocation." He realized that a machine sitting idle was not "free." It was a liability.

We are currently in the 1887 of the digital age.

Most companies today treat AI like a utility bill. They pay a flat fee to a provider and hope their operations get "more productive." They have no idea what it costs to generate a feature, no idea what the ROI of a specific agentic session is, and no idea how much of their "AI Spend" is being evaporated by idle chat windows and hallucinated loops.

This document is the announcement of a new discipline: Compute Accounting.

I. The Primitive Protocol: Engineering the Runtime

The era of software-as-a-product is ending. We are entering the era of Runtime-as-an-Outcome.

For thirty years, the software industry has operated on the "Managed Experience" model. You bought an interface, a database, and a workflow engine, all wrapped in a brand name. This model was a necessity when the "Primitive" (the raw compute) was too difficult for humans to interact with directly.

But as AI agents have advanced, the "Experience" has become a bottleneck.

A modern agentic system is not a tool; it is a factory. It consumes electricity, GPU cycles, and human intent. To manage this effectively, we must move beyond the opaque subscription and down to the Primitive:

  1. RTC (Runtime Credits): Raw compute seconds on a dedicated VM.
  2. MSG (Message Credits): Orchestration signals between agents.
  3. TMC (Taskmaster Credits): Strategic reasoning and multi-step planning.

This is the first step toward Direct Execution. We are stripping away the "SaaS Wrapper" and giving you the raw primitives of the machine. By managing the runtime directly, you are no longer paying for the existence of a tool; you are paying for the execution of a task.

II. The History of Modular Revolutions

Modularity is the precursor to every period of explosive economic growth in human history.

In the early 19th century, every machine was unique. If a bolt broke on a steam engine, you had to find a blacksmith to hand-forge a new one. This was the "Integrated Era."

Then came Eli Whitney and the concept of Interchangeable Parts. By standardizing the primitive (the part), Whitney allowed for the mass production of the outcome (the machine).

The software industry is currently in the pre-Whitney era. Most SaaS products are bespoke, integrated monoliths. If you want a CRM, you must buy the whole CRM, even if you only need the "Lead Generation" outcome.

The Outcome Engineering Framework (OEF) is the "Interchangeable Parts" moment for software. We have unbundled the monolith into four raw primitives:

  • Ingestion (Firecrawl): The ability to see the world.
  • Reasoning (Claude/GPT): The ability to think about the world.
  • Execution (Sandbox): The ability to change the world.
  • Persistence (GitHub): The ability to remember the world.

When these are modular, you don't buy a product. You architect a pipeline.

III. The Unit Economics of Intent

In 1923, Donaldson Brown of General Motors developed the ROI (Return on Investment) formula. He realized that to manage a giant like GM, you couldn't just look at sales; you had to look at how much capital was being deployed to get those sales.

In the digital era, Intent is Capital.

Every time you prompt an agent, you are deploying human intent into a compute environment. If that environment is a "Managed Chatbox," you have no way to measure the efficiency of that deployment.

Compute Accounting allows us to perform Variance Analysis on Intent:

OutcomePredicted RuntimeActual RuntimeVarianceEfficiency
Build Navbar45 Seconds120 Seconds+75sLow (Agent Loop)
Fix Auth Bug10 Minutes2 Minutes-8mHigh (Direct Hit)
Deploy Landing Page5 Minutes6 Minutes+1sOptimal

This is not "monitoring." It is Performance Management. We can now see which models are cost-effective for specific tasks, which prompts are leading to "Compute Waste," and which engineering outcomes are actually profitable.

IV. The 5-Minute Window: The Physical Reality of Readiness

One of the most controversial decisions in the OEF was the Billed Warm Window.

In the old world of software, infrastructure costs were hidden or subsidized. This created a belief that "Idle" was free.

It is not. A VM sitting in a "Warm" state is a high-performance machine ready to respond in seconds. That readiness has a physical cost in the data center. Vercel and other infrastructure providers charge for this readiness because the CPU cycles are reserved and the memory is allocated.

By billing for the 5-minute warm window, we are introducing Truth in Accounting.

  • Scenario A: You send a message, the agent works, and you close the tab. The VM stays warm for 5 minutes. You are billed for those 5 minutes because we are holding a dedicated engineering environment open for you.
  • Scenario B: You send a message, the agent works, and you click "Stop Sandbox." The VM is reclaimed. The billing stops immediately.

This choice forces the operator to be an Active Manager of Resources. It eliminates the "Ghost in the Ledger", the invisible waste that plagues corporate AI accounts.

V. The Lego Model: Democratizing the Primitive

The legacy platforms (Salesforce, ServiceNow, Microsoft) want to sell you an "Experience." They want to be the "Operating System for Your Business."

This is a play for Institutional Moat. If they own the experience, they own the data and they own the access. Ben Thompson (Stratechery) describes this as "Integration vs. Modularity." The giants integrate to capture value; we modularize to distribute it.

The "Lego Model" is the nuclear option for this monopoly.

By giving you modular access to the primitives, we are allowing you to route around the legacy giants. You don't need a "Managed Platform" if you can assemble a "Lead Generation Pipeline" from raw primitives:

  • A Firecrawl primitive to ingest data.
  • A Claude primitive to reason.
  • A Sandbox primitive to execute.
  • A GitHub primitive to persist.

You aren't a user of our software. You are an Architect of the machine.

VI. The Primitive in Practice: Real-World Outcome Scenarios

To understand the power of the primitive, we must look at how it replaces the "Managed Experience" in high-stakes business environments.

Scenario A: The Just-In-Time Supply Chain

  • The Managed Trap: A company pays $200,000/year for a supply chain visibility platform. The platform shows them a dashboard of delayed shipments. A human manager then has to manually call vendors to resolve the issues.
  • The Primitive Solution: An OEF pipeline is assembled. A Firecrawl primitive monitors port activity and vessel GPS in real-time. A Claude primitive reasons on the impact of delays. A Sandbox primitive executes automated re-routing orders via the vendor's API.
  • The Result: The "Visibility" outcome is replaced by the "Resolution" outcome. The company pays only for the Runtime required to fix the delay.

Scenario B: The Continuous Legal Audit

  • The Managed Trap: A financial firm hires a Big 4 firm for a quarterly compliance audit. The audit is a snapshot in time, cost-intensive, and prone to human error.
  • The Primitive Solution: A GitHub primitive stores the firm's internal policies. A Firecrawl primitive ingests daily regulatory updates from the SEC and FINRA. A Claude primitive performs a "Difference Analysis" between the new laws and the internal code. A Sandbox primitive generates a Pull Request with the necessary policy updates.
  • The Result: The firm moves from "Quarterly Snapshots" to Real-Time Compliance. The cost moves from a $50,000 engagement to a $5.00 RTC charge.

Scenario C: The Lead Generation Factory

  • The Managed Trap: A startup pays for a $1,500/month CRM and a $500/month lead database. Their SDRs spend 4 hours a day manually searching for leads and copy-pasting info into emails.
  • The Primitive Solution: A Firecrawl primitive is scheduled to scan industry news for specific triggers (e.g., new funding rounds). A Claude primitive extracts the decision-maker's LinkedIn and recent public statements. A Sandbox primitive executes a personalized outreach sequence via the mail API.
  • The Result: The "Management of Leads" is replaced by the "Generation of Outcomes." The startup pays only for the compute seconds required to find and contact a specific target.

Scenario D: The Self-Healing App

  • The Managed Trap: An enterprise pays for an APM (Application Performance Monitoring) tool that pings an SRE (Site Reliability Engineer) when a bug occurs. The SRE wakes up at 3:00 AM, searches the logs, and pushes a fix.
  • The Primitive Solution: A Sandbox primitive runs a continuous monitoring agent. When an error is detected, the agent autonomously retrieves the log from GitHub, reasons on the fix using Claude, and executes a "Diagnostic Deploy" in a isolated environment. Once verified, it pushes the fix to production.
  • The Result: The "Monitoring" outcome is replaced by the "Resilience" outcome. The cost of a human "On-Call" rotation is replaced by cents of RTC.

Scenario E: The Autonomous Content Strategist

  • The Managed Trap: A brand pays $5,000/month to a social media agency. The agency spends 2 weeks planning a "Content Calendar" that is often out of sync with real-time trends by the time it's published.
  • The Primitive Solution: A Firecrawl primitive tracks trending topics in the brand's niche every hour. A Claude primitive reasons on how the brand can uniquely contribute to the conversation. A Sandbox primitive generates the creative assets and schedules the posts.
  • The Result: The "Management of Channels" is replaced by the "Capture of Attention." The brand pays only for the compute required to be relevant in the moment.

Scenario F: The Real-Time Financial Arbitrage

  • The Managed Trap: A hedge fund hires junior analysts to read earnings transcripts and news feeds. The delay between an event and an investment decision is measured in hours or days.
  • The Primitive Solution: A Firecrawl primitive ingests every SEC filing and earnings call transcript within seconds of release. A Claude primitive performs a multi-model sentiment and risk analysis. A Sandbox primitive executes the trade strategy based on the reasoning.
  • The Result: The "Analysis of History" is replaced by the "Execution of Insight." The fund pays only for the RTC required to win the trade.

Performance Comparison: Managed vs. Modular

MetricManaged Platform (SaaS)Modular Primitive (OEF)
Unit of CostPer User / Per MonthPer Outcome / Per Second
Waste ProfileHigh (Idle Subscriptions)Zero (Direct Execution)
AdaptabilityRigid (Vendor Roadmap)Infinite (Lego Re-assembly)
LatencyHuman-in-the-LoopAgent-Direct
AccountabilityOpaque (Black Box AI)Transparent (Ledger-Backed)
Outcome DensityLow (Value Diluted by Seats)High (Pure Value vs. Runtime)
Outcome VelocityLow (Throttled by Human Speed)High (Limited Only by Compute)

VII. The Historical Parallel: The Unbundling of the Cloud Bell

In 1982, the United States government broke up the Bell System (AT&T). For decades, Bell had owned everything: the lines, the switches, and even the physical telephone in your house. You couldn't buy your own phone; you had to lease it from Bell.

This was the "Integrated Era" of telecommunications. It was efficient for the provider but stifling for the world.

When the "Cloud Bell" (Salesforce, Microsoft, ServiceNow) was broken, it didn't happen in a courtroom. It happened in the codebase. By unbundling the Primitive (the line and the switch) from the Experience (the physical phone), the world saw the birth of the modem, the internet, and the mobile revolution.

We are currently witnessing the Unbundling of the Cloud Bell.

The legacy giants want to lease you the "Digital Telephone" (the SaaS license). We are giving you the lines and the switches. When you have the primitives, you no longer need the permission of the giant to build the outcome.

VIII. Cites, Sources, and the 150-Year Precedent

Compute Accounting is the logical conclusion of 150 years of management theory.

  1. The Toyota Production System (TPS): Taiichi Ohno identified "Muda" (Waste). In digital work, the ultimate Muda is Idle Runtime. By providing a 5-minute warm window and a "Stop" button, we are implementing Just-In-Time (JIT) compute.
  2. The Theory of Constraints (Eliyahu Goldratt): The bottleneck is no longer "Coding Speed"; it is "Verification Speed." Our Timebox system focuses the agent's life on the outcome, forcing it to work within the constraints of the budget.
  3. Gartner Cloud Data (2024): Gartner reports that 30% of cloud spend is wasted due to over-provisioning and idle resources. Compute Accounting is the first framework designed to eliminate this at the agentic layer.
  4. Brooks's Law (1975): "Adding manpower to a late software project makes it later." Compute Accounting proves that Runtime Orchestration is the critical path. A single agent with a 10-hour Timebox is often more efficient than multiple agents with fragmented Timeboxes due to coordination overhead.
  5. Direct Manipulation (Ben Shneiderman): The OEF focuses on Tactile Performance. Whether on a workstation or a mobile device, the operator has direct control over the compute environment. Dragging a Timebox with a finger isn't just a UI feature; it is the physical act of managing a digital factory.

VII. The Final Gloat: The Collapse of Subscription Idle

The technology industry is transitioning.

As operations move toward Runtime and away from static subscriptions, the valuations of legacy giants built on "Subscription Idle" will be questioned. They depend on the "Unused Resource", the license that is paid for but never utilized.

We have no interest in the unused resource. We have no interest in "Engagement Metrics." We are interested in Outcome Density.

We are giving you the legos. We are giving you the primitives. We are giving you the ledger.

The "Work Lie" was that you needed a manager to watch your time. The truth is you need a Timebox to manage your compute.

Build the Diamond.

Sources & Economic Data

Filed under: Compute Accounting · Modular Computing · Outcome Engineering · Performance Management · The Lego Model · Brighton Mlambo · Management Accounting · OEF

Written: April 28, 2026

The Modular Primitive - OutcomeDev Blog