// RESEARCH APPENDIX
Every Claim,
Cited and Transparent
The quantitative claims in our thesis are based on specific formulas, peer-reviewed research, industry reports, and clearly stated assumptions. This page provides full transparency on our methodology.
// FORMULA_01
Communication Overhead Formula
For n people in a fully-connected team:
Communication channels = n(n-1)/2
Source: Brooks, The Mythical Man-Month (1975)
| Team Size | Communication Channels | Implication |
|---|---|---|
| 7 people (CTO model) | 21 channels | Manageable without formal process |
| 50 people | 1,225 channels | Requires hierarchy and managers |
| 100 people | 4,950 channels | Requires bureaucracy, PMO, Jira, Confluence |
| 200 people | 19,900 channels | Most time spent coordinating, not building |
This formula explains why teams naturally introduce managers, processes, and documentation as they grow. The overhead isn't optional — it's mathematically inevitable beyond ~7 people. Miller's Law (1956) independently confirms this threshold.
// FORMULA_02
Brooks's Law, Revisited
The Original Law (1975)
"Adding people to a late software project makes it later."
— Fred Brooks, The Mythical Man-Month
The Inverse (2025)
"Removing people from a software project (and replacing with AI) makes it faster."
— The CTO Thesis
Why the Inverse Now Holds
- *Brooks's Law was based on communication overhead — each new person adds n-1 new communication paths
- *AI doesn't add communication paths — it amplifies existing human capacity without coordination cost
- *Replacing 10 people with 1 CTO + AI removes 45 communication channels while retaining most execution capacity
- *DORA 2025 confirms: AI improved individual metrics but organisational delivery stayed flat — structure, not tools, is the bottleneck
// ESTIMATES
AI Force Multiplication Estimates
Based on 2025-2026 capabilities and our direct experience building with AI tools. Cross-referenced with Google and Anthropic internal studies showing 1.5-2x gains for senior engineers.
| Task Type | Multiplier | Notes |
|---|---|---|
| Boilerplate code | 3-5x | CRUD, scaffolding, config — heavily accelerated by AI (Google internal data) |
| Standard CRUD | 2-3x | Endpoints, forms, data flows (Anthropic internal studies) |
| Complex business logic | 1.5-2x | Requires CTO judgment; AI accelerates but does not replace expertise |
| Novel architecture | 1-1.5x | AI assists with patterns; METR found 19% slowdown on unfamiliar complex tasks |
| Strategic decisions | 1x | No AI advantage — pure human judgment |
Weighted Average
For typical product development: 1.5-3x acceleration (varies significantly by task type)
METR study caveat: experienced devs were 19% slower with AI on complex, unfamiliar tasks. The multiplier only holds when senior practitioners direct the work — which is exactly what our model provides.
// VALUE_CREATION
Value-Creation Percentage Estimates
The percentages used to calculate "effective output" are based on time-tracking studies, industry surveys, and our direct experience across dozens of engagements.
Traditional Org Breakdown
- CTO:70% coordination (board meetings, stakeholder management, strategy alignment) → 30% value-creation
- VP Eng:85% coordination (team management, planning, reporting) → 15% value-creation
- Eng Manager:90% coordination (1:1s, standups, planning, performance) → 10% value-creation
- Senior Eng:50% coordination (code review, mentoring, meetings) → 50% value-creation
- Engineer:40% coordination (meetings, waiting, context switching) → 60% value-creation
CTO + AI Model Breakdown
CTO with AI: 15% coordination (minimal sync meetings) → 85% value-creation
The 85% figure is achievable because there's no management layer, no handoffs, and no waiting for others. McKinsey's "human-agent delivery pods" model describes exactly this structure.
// ASSUMPTIONS
Salary & Cost Assumptions
All salary figures are UK market rates for technology companies (total compensation including base + bonus). These are the figures used in our thesis comparison models.
| Role | Total Compensation | Source |
|---|---|---|
| CTO | £280k | Glassdoor, Levels.fyi 2025-2026 UK data |
| VP Engineering | £220k | Glassdoor, Levels.fyi 2025-2026 UK data |
| Engineering Manager | £160k | Glassdoor, Levels.fyi 2025-2026 UK data |
| Senior Engineer | £140k | Glassdoor, Levels.fyi 2025-2026 UK data |
| Engineer | £110k | Glassdoor, Levels.fyi 2025-2026 UK data |
Additional Cost Factors
- *Benefits & overhead: Add 25-35% to base salary for true employer cost (NI, pension, benefits)
- *AI tooling: ~£1k-2k per developer per year (Cursor, GitHub Copilot, Claude, etc.)
- *Infrastructure: Not included — assumed similar for both models
- *Recruiting costs: 15-25% of first-year salary (not included in annual figures)
// BIBLIOGRAPHY
Sources & References
Every major claim in the thesis is backed by the following sources. Where we extrapolate beyond the data, we say so explicitly.
McKinsey & Company, December 2025
"Superagency in the workplace: Empowering people to unlock AI's full potential"
- —80% of C-suite executives already running agentic AI pilots
- —20-30% expected contraction in traditional tech services market
- —$400B incremental AI spending opportunity by 2030
- —15-30% of current roles' work could be done by agents within 3 years
- —Winning model identified as "human-agent delivery pods with centralized governance"
Fortune, February 2026
"AI's 'February 2020' Moment"
- —Up to 50% of entry-level white-collar jobs at risk in coming years (Amodei prediction, reported by Fortune)
- —Matt Shumer (CEO HyperWrite): "I describe what I want in natural language and it just appears"
- —Disruption is visible but most organisations haven't felt the impact yet
Citrini Research, 2025
"The Intelligence Displacement Spiral"
- —Agentic coding tools could replicate mid-market SaaS in weeks (scenario analysis)
- —Global intelligence crisis projected by 2028, software as ground zero
- —In a displacement scenario, tech ZIP code housing could drop 8-11% (hypothetical analysis)
- —PE-backed software companies face rising default rates
- —Self-reinforcing cycle: cost savings fund more AI investment, accelerating displacement
DORA State of DevOps Report, 2025
"Accelerate: State of DevOps"
- —AI improved individual developer metrics (lines of code, PR speed)
- —Organisational delivery throughput remained flat — tools alone don't solve structural problems
- —Teams that redesigned workflows saw genuine gains; teams that added AI to existing processes did not
METR (Model Evaluation & Threat Research), 2025
"Measuring the Impact of AI Coding Tools on Developer Productivity"
- —Experienced developers were 19% slower with AI on complex, unfamiliar tasks
- —AI tools helped most on simple, repetitive tasks — least on novel or architectural work
- —Conclusion: seniority and judgment matter more than tooling alone
Google & Anthropic Internal Studies, 2024-2025
Internal productivity measurements
- —1.5-2x real productivity gain for senior engineers using AI tools
- —Gains concentrated in developers who could already write the code — AI accelerated, not replaced, judgment
Dario Amodei, CEO Anthropic, 2025
"Machines of Loving Grace"
- —AI will match expert-level capability "across virtually every relevant field"
- —Timeline for workforce restructuring is compressing faster than consensus estimates
Fred Brooks, 1975
"The Mythical Man-Month"
- —"Adding people to a late software project makes it later"
- —Communication overhead grows as n(n-1)/2 — the mathematical basis for our coordination cost model
George A. Miller, 1956
"The Magical Number Seven, Plus or Minus Two"
- —Humans can effectively coordinate with 7±2 people without formal hierarchy
- —Beyond this threshold, teams require managers, processes, and documentation — all overhead
// CAVEATS
Limitations & Caveats
We believe in intellectual honesty. Here's what we're less certain about.
- 01Sample size: Our direct experience is with dozens of projects, not hundreds. Individual results vary based on domain, complexity, and client engagement.
- 02Selection bias: CTOs who self-select into this model may be unusually high-performing. The model requires a specific kind of practitioner.
- 03AI capabilities are a moving target: These estimates reflect 2025-2026 AI. They will likely be conservative within 12 months as models improve.
- 04Domain-specific: Results are strongest in web/mobile SaaS. Hardware, embedded systems, and heavily regulated industries have different dynamics.
- 05Coordination percentages: Based on industry surveys, time-tracking studies, and our direct experience — not controlled experiments.
- 06Third-party research: McKinsey, Citrini, and Fortune figures are cited as reported. We have not independently verified their underlying methodologies.
Questions About Our Methodology?
We're happy to discuss our research and assumptions in detail. Book a call or read the full thesis.