What a teamwould have cost.

The talk’s result priced twice: as a Swiss delivery team, and as it happened with agentic AI. The owner’s hours count in both.

Counts are from git as of , like every figure in the talk. Estimates are marked, stated with their basis, and can be changed below.

Price labor as
Team estimate
Conventional team estimated FTE on average for months, plus the owner
Agentic AI, as it happenedOne person and agents, about months, owner hours
Difference× cheaper less, about × faster

What was built

Both bills buy the same result: two products, CompanyGraph and GuestGraph, and this site on a model of its own. Every figure here is counted.

public repositories
commits on default branches
merged pull requests
releases
of the commits written with Claude
lines of code
lines of tests
lines of Markdown
specs
decisions recorded in them

The two bills

One scale for both. The owner’s hours are the same color in both, because they are paid in both.

The owner’s hoursThe delivery teamTools, AI and hosting
Conventional
Agentic AI
LineConventionalAgentic AI
The owner’s hours h h
The delivery team person-months—
Tools, AI and hostinglicensesas paid

Time, and the owner’s time

The team takes longer from start to live. The owner’s hours barely change, which is the talk’s point: the time building saved went into deciding.

Conventional
h
Agentic AI
h

The team it would take

A Swiss team at market rates, available from the first day. Its person-months are estimated per workstream below; the conservative case matches the measured code at about lines of tested code per person-month.

RoleFTEPerson-monthsCHF an hourCost
Tech lead and solution architect
Senior backend engineers
Senior frontend engineers
UX and UI designer
DevOps and platform engineer
QA and test engineer
Business analyst and knowledge modeler
Technical writer and translator
Delivery manager
Owner and lead, at a lead architect’s market rate in both
The delivery team

Where the person-months go

GuestGraph engine and Apaleo connector
Design system
Three bilingual sites and the talks
Meta-model, command line and checks
Obsidian plugin and its end-to-end tests
Quality assurance
Delivery management
MCP server on three hosts
Conventions, CI and the cloud platform
Chat server on three hosts
Three models, as content

What it actually took

One person with Claude Code, from the first commit on to . The hours are the days git shows work, times the hours a day the owner gives.

LineBasisCHF
The owner, from days with commits × h
The owner, before days with commits × h
Claude subscription as paidMax 5x until mid-June, then Max 20x: USD , plus Swiss VAT
Narration voice as paidUSD of ElevenLabs credit
The chats’ API as paidUSD of Anthropic credit for Sonnet 5, plus VAT
Google Cloud hosting countedSix services that scale to zero, inside the free tier
Two domains estimatedcompanygraph.io and guestgraph.io for a year
Agentic total

The AI was the smallest line: of the total. At the API’s list price instead of the subscription, USD , the total would be , still × cheaper. The cost moved to the owner’s hours, which is where the deciding is.

Would both approaches produce the same result?

Not quite. A good Swiss team could match each bar here, spec by spec and test by test. At this pace it would deliver less scope and write down less of its reasoning, and it would bring what one person with agents lacks: independent review, real users and shared ownership. The evidence is counted; the rating is the owner’s judgment.

DimensionEvidence from this buildCould a team do the same?Outcome
Specification specs and plans, written before the code, with decisions each traced to a quote; revised later and dropped, so they were used.Yes. The decisions are the owner’s either way, but writing them down at this depth costs an architect weeks a quarter, and most teams cut it first.Agentic ahead
Implementation revert in commits, and pull requests closed without merging: the gate turned work away.Yes. Senior engineers write code like this and bring judgment from years in production, which an agent has only when the owner supplies it.Even
Tests lines of tests for every ten of code; the plugin tested end to end in a real Obsidian; CI on every change.Yes, and the team includes a QA engineer for it. Under a deadline, tests are the second thing a team trims.Agentic ahead
ConsistencyOne set of conventions in every repository, rolled out in release waves.Hard. Style and wording drift between people; linting gets a team close.Agentic ahead
Scope and iterationIdeas prototyped and declined within a day; two products and three models in about fifteen weeks.A team runs fewer experiments, since each costs a sprint, and scopes smaller. Often healthy, but a different product.Agentic ahead
German and EnglishEvery page in both, with a translator’s review and a glossary.Yes. A professional translator is the standard for Swiss Standard German.Even
Independent reviewThe owner is the only human reviewer; agents review agents, and the owner merges.Better: two people on every change, and security review by someone who did not write it.Team ahead
Design with usersOne person’s judgment, refined in review; no usability or accessibility study with real users.Better: a designer with user research sees what the builder cannot.Team ahead
OwnershipOne person holds the context; the specs write the reasoning down, which lowers the risk without removing it.Better: the knowledge is spread across the team.Team ahead
OperationsEvery host deploys through CI; there is no on-call and no service level.Better with a DevOps engineer. Neither side’s running costs are counted.Team ahead

Closing the gaps where the team is ahead — an external code and security review, a usability and accessibility study, a month to bring in a second maintainer — adds about . The agentic total would then be , still × cheaper.

These are counts from git, not an audit. Nobody outside the project has reviewed the code, so the implementation row compares capability, not a measured outcome.

How the figures are made

Employer cost is the salary × for social charges and workplace, over productive hours a year. Salaries sit at the Zurich market middle, a senior engineer at against a reported average of about CHF 137,000 (Levels.fyi, Glassdoor).

Contractor rates are CHF – an hour, inside the reported day rates of CHF 850–1,500 for Swiss IT contractors (salary2freelance).

The owner is priced as a lead architect at market rate, on the employer basis or CHF an hour, not at what he earns or charges.

Dollars are converted at , near the 2026 average. Anthropic bills Switzerland in dollars, plus VAT.

What this leaves out

Hiring the team, which in Switzerland usually takes three to six months before the months above begin. The meetings and handovers a larger team needs. Running costs on either side. The risk the talk names: the speed rests on a few providers’ prices and terms.