Proof
Three AI products, built by a senior lead directing AI.
Three products and 27 AI capabilities, each one named. Scope and working days are counted from each product's git history, the record of every change to its code, as of 2 July 2026. CyberSee is live for you to try.
PropTech · Client engagement
A market-intelligence platform for letting agents
Letting agents had no way to see their own market. The information existed, scattered across thousands of public sources, and assembling it by hand was never worth anyone's time. This platform collects it continuously, reconciles it into a single dataset, grades it with AI, and turns it into something an agent can act on the same morning.
Client build, pre-launch. Delivered end to end in one funded stage: a public site, a multi-source data engine, and an AI proof of concept.
8 AI capabilities, built in 68 working days by a senior lead directing AI.
- Grades listing photos A to F with a vision model
- Picks out the features in each photo
- Recognises which room a photo shows
- Flags watermarked photos
- Scores photo quality
- Estimates floor area and checks it against EPC records
- Turns portfolio data into upgrade recommendations
- Reads market news and classifies its sentiment
Counted from git
65,395 lines of production code
9,532 lines of tests
19 application pages
828 commits
68 working days
Trunk commit cadence: median 13 minutes between commits, with deploy-on-merge CI.
Two more, built the same way
Different domains, measured the same way.
Cyber security · SaaS
CyberSee
Getting certified has meant hiring a consultancy and losing most of a year to it. CyberSee runs that path in software instead: posture assessment, the evidence a certifier asks for, and guidance on what to fix next, kept current rather than assembled in a panic the week before an audit.
Live. Sign up free and look around it yourself. cybersee.ai
10 AI capabilities, built in 82 working days by a senior lead directing AI.
- Answers questions from retrieved sources
- Drafts policies from a company profile
- Drafts how each control is implemented
- Tailors controls to the company
- Suggests risks
- Works out which controls apply
- Drafts the scope statement
- Builds a company profile from its website
- Analyses the policies a company already has
- Swaps sensitive data for tokens before it reaches the model
Counted from git
53 pages
205 endpoints
107,257 production LOC
11,820 test LOC
10 integrations
Legal AI · Platform
An AI legal-workflow platform
Legal work is expensive because it is slow, and slow because somebody has to read everything. This platform breaks a matter into the issues that will decide it, finds the law governing each one, and shows its working, so a lawyer can check it rather than take it on trust.
Private pre-beta, access by invitation.
9 AI capabilities, built in 33 working days by a senior lead directing AI.
- A ten-step adversarial case analysis
- Breaks each matter into issues and sub-issues, with structured tool use
- AI agents that challenge each other's arguments in parallel
- An assessment of each issue with a stated confidence
- Estimates given as low, mid and high ranges
- An analysis of likely costs
- Retrieves relevant case law
- Benchmarked against cases with known outcomes, by Brier score and accuracy
- Drafts a structured report
Counted from git
17 pages
51 endpoints
23,531 production LOC
8 integrations
For scale, a traditional team typically costs 4-6x our fixed price for the same scope (our estimate). How we count
How this is possible
A senior lead directing AI, with specialists pulled in where scope needs them, produces what used to take a coordinated department.
- AI writes the code, tests and documentation. The lead directs and reviews, applying 20+ years of engineering judgment to every decision.
- Specialists join for the work that needs them, design, security, data, under the same single accountability.
- Builds start on infrastructure patterns already proven in production, so work moves on day one.
- Fixed price against a signed done-test, with the delivery risk on us.
How we measure
Code, tests, pages, commits, cadence and working days are counted from each product's git history and infrastructure by our measure tool, as of 2 July 2026. AI capabilities are listed by name. The traditional-team comparison is our estimate.
The traditional-team benchmark prices the counted scope component by component against published market rates, each row sourced and graded for confidence. Set against our fixed price for the same scope, that is typically 4-6 times. For ventures we believe in, we can share in the upside instead of charging full price.
It excludes recruitment, management overhead and the 52% of projects the industry reports as challenged or failed, all of which a traditional team pays.
Working days are git-counted; days spent across multiple projects are counted in full against each, so correcting the overlap only widens the gap.
AI API spend, third-party vendor spend, and documentation effort are tracked separately and are not included in these figures.
Want these numbers for your own codebase?
The Codebase X-Ray runs this tool over your code and reports what it would cost a traditional team to build. Free, and yours whatever you decide next.
Want this level of evidence on your own system?
Start with a fixed-price assessment, or tell us what you're building and we'll come back with a scope and a fixed price against a signed done-test.