People sometimes hear our anti-extraction stance and assume it's a moral stance dressed up as a business stance. It isn't. Refusing extraction patterns is the most boring kind of strategy: the kind that compounds.
The argument in one paragraph
Extraction patterns — seat licensing as the deliverable, opaque pipelines, vendor lock, retainer treadmills — generate revenue this quarter and destroy the thing that generates revenue every quarter after that. The thing they destroy is trust, and trust is the substrate that referrals and renewals grow on. Trust compounds. Extracted dollars do not. A book of business built on extraction looks great on a Q3 dashboard and looks like a leaky bucket on a five-year P&L. We are playing the five-year game on purpose.
Why this isn't a moral argument
A moral argument would be "extraction is wrong, don't do it." That's true, but it isn't load-bearing here. The load-bearing claim is mechanical, and it's about how clients actually behave after an engagement ends.
A client who walks away with runnable artifacts, a written runbook, a complete cost map, and a clear sense of what they bought (Class C — configuration of the engagement itself, observable in the signed scope and the artifact list) will do three things over the next eighteen months:
- Refer at least one peer, because the artifact survives contact with their operating reality and other operators can see that.
- Come back for a second engagement at a higher scope of trust, because the first engagement made them more capable, not more dependent.
- Tolerate honest bad news from us if the second engagement gets hard, because the first one taught them what our bad news looks like.
A client who walks away with a per-seat subscription they don't understand and a vendor inbox they can't escape from will do none of those three things. They will churn quietly, badmouth the category, and warn the next operator off it. We've watched this happen across the AI services market for three years.
Refusing extraction is therefore the cheapest, most durable form of marketing we have access to. It is pro-compounding, not anti-revenue.
The tiering point
Anti-extraction also forces a question most agencies refuse to answer in writing: what tier of work does this client actually need? Themesis has written usefully on this — her piece on focus for AI workers sorts AI practitioners into three career tiers and argues each tier needs a different depth of study to do honest work (Class E — expert framing we find load-bearing). We use the same shape on the client side: an engagement priced at innovator-tier when the client needs tradesperson-tier work is itself an extraction pattern, just dressed in better fonts.
The honest move is to scope to the tier. Sometimes that means a smaller engagement than the prospect arrived expecting. We lose those proposals on price and win them on the second call, six months later, after they've watched a competitor extract them on a project that should have been a workshop.
What pro-compounding actually looks like in the contract
The clauses that make a contract anti-extractive are short and unambiguous. The client owns every artifact. Every external dependency is listed with its monthly cost and a substitution note. The runbook is tested by an outsider before sign-off. If the engagement is going badly, we are required to say so in writing within a week of noticing — with the falsifier (Class C, again — these are configuration commitments, not slogans).
None of those clauses cost us money on a healthy engagement. All of them save the client money on an unhealthy one. The asymmetry is the whole point.
The strategic claim, plainly
We are not refusing extraction because extraction is unkind, though it is. We are refusing it because the math of a referral-driven services book makes extraction a slow-motion balance-sheet fire, and we would like our book of business to still be here in 2031.
If you want to see what anti-extraction looks like in a signed scope, the place to start is the engagement model itself.
- See the four counter-commitments in plain English: Anti-Extraction Engagement Model
- The contract language that makes those commitments real: Partnership vs Vendor Contract
- Bring a real problem and let us scope to the tier: Workshop