Most AI agency proposals describe their team as a shape rather than a list of people. “Three senior engineers, one ML lead, one technical PM” is the standard form. The shape is impressive on a slide and meaningless under a deadline. Whoever signs the contract is paying for a configuration of seats, not for the specific humans who will write the prompts, choose the model, debug the eval regression at 11 p.m. On the day before launch, and; critically; own the system on the morning after the agency rolls off. The buyer who accepts the shape inherits the agency’s staffing flexibility as a personal liability, and discovers in week 5 that the engineer described in the pitch as “lead” is splitting attention across three other projects while a contractor they have rarely met is writing the retrieval layer.
This piece argues that AI proposals in 2026 should require named individuals, with GitHub profiles, links to specific prior PRs, and a non-substitution clause that makes swaps a contract event rather than an internal staffing decision. I cover why named-individual proposals raise the floor for both sides, what the agency loses (the answer is genuinely less than agencies fear), what the buyer gains (continuity, leverage, faster onboarding), what good clauses look like, and the narrow set of conditions under which substitution should be acceptable. The frame for many of it is the AI agency manifesto, which holds that 2026 buyers are paying for shipped, eval-gated software written by specific operators rather than for an org chart.
Decision Scope
This article is an editorial decision framework, not legal, financial, security, or accounting advice. Treat numeric examples as illustrative planning heuristics unless a source is cited, then validate the assumptions against your own contracts, data, controls, and budget model before acting.
Why team-shape proposals became the norm
Three forces produced the role-based proposal. The first is hiring volatility; agencies that staffed up during the 2023–2024 LLM rush could not promise specific people in February when they were not sure who would still be at the company in April. The second is utilization economics: an agency that names an engineer in a proposal commits to scheduling that engineer’s calendar against the project, which constrains the agency’s ability to move bodies between accounts and crushes utilization rates. The third is sales mechanics. A pitch built around a generic ML lead is reusable across ten prospects in a quarter; a pitch built around a specific engineer’s GitHub profile and shipped systems is reusable across one. The role-based proposal optimizes for the agency’s cash flow, not for the buyer’s outcome.
The buyer-side consequence is that the proposal review measures the wrong thing. The buyer reads a paragraph about an “ML lead with 8 years of experience and a Stanford CS background” and feels reassured. They do not learn whether the actual person on the project has shipped a retrieval-augmented system at the relevant scale, has debugged a prompt-injection incident in production, has written a single test for an LLM call, or is the same person who wrote the system the agency is showing as case study #2. The role description is performative; the named-individual disclosure is operational. AI engineering, more than most disciplines, lives or dies on the specific muscle memory of the specific human at the keyboard.
What named-individual proposals require
A named-individual proposal lists the people, not the seats, and attaches three artifacts to each name. The first is a GitHub or equivalent profile; public if available, a curated portfolio of repositories under NDA if not. The second is a short list of prior PRs or shipped projects with links and a one-paragraph description of what the engineer specifically owned. The third is a percentage allocation expressed as a number (“50% of David’s calendar from week 1 through week 8”) with a defined floor under which the engagement triggers a renegotiation clause.
The proposal also identifies a primary engineer per workstream. On a typical 12-week LLM build with retrieval, agentic tool use, and an evaluation pipeline, that is three names. Each named engineer has a phone number on the runbook, an account on the client’s observability stack, and authorship on the PRs that will eventually need to be defended in a postmortem. The pattern is borrowed from architecture firms and surgical teams, where it has been load-bearing for decades, and from the engineering-led agencies that have been quietly running it on AI work since 2023.
The non-substitution clause is the legal backbone. It says: substitution of any named engineer requires written notice to the client at least 14 days in advance, identification of the proposed replacement with the same artifacts as the original, a 1-week shadow period during which both engineers are billing at half rate, and the right of the client to terminate the affected workstream at no fee if the substitution is unacceptable. The clause has narrow exceptions; illness, departure, or legal cause; handled separately below. Without the clause, naming engineers is decoration; with it, naming engineers is a commitment.
Why agencies fear this; and why the fear is mostly misplaced
The agency objection runs as follows: naming engineers locks in our staffing, kills utilization, makes vacation scheduling impossible, and exposes us to the worst case where a key engineer quits mid-project and we owe the client a refund or a credit. Each piece of the objection is a real operational concern, and each one is solvable with proposal hygiene the agency should have anyway.
Lock-in is the smallest of the problems. The agency that has named three engineers on one project still has flexibility on the other twenty projects, and the staffing constraint is the same constraint a law firm or an architecture firm operates under without obvious distress. Utilization is real but recoverable: the agency that names engineers can charge for them at higher rates because the buyer is buying specific people rather than fungible bodies, and the rate increase typically dwarfs the utilization hit. Vacation is a paperwork problem solved by scheduling shadow weeks at the front of the project. The mid-project departure is the genuinely hard case and is the one the non-substitution clause’s “departure” exception is designed for; the agency notifies, names a replacement, and absorbs the shadow week’s cost as part of the retention risk it should be pricing into the deal in any case.
What the agency loses, honestly, is the ability to bait-and-switch. The proposal that pitches the founder and delivers the junior, the case study built on the work of a senior who has since left, the “team of 3” that turns out to be one staff engineer split across four projects; those patterns are no longer available. Most reputable agencies stopped doing them years ago and will not feel the loss. The agencies that genuinely depend on bait-and-switch will fail this clause and, by failing, will reveal their economics to the buyer. That is the clause working correctly.
What the buyer gains; beyond the obvious
The visible win is continuity: the engineer who designed the retrieval layer in week 2 is the one who debugs it in week 11, and the muscle memory of the system stays in one head rather than being lost in a handoff. The deeper wins are leverage and onboarding speed.
Leverage shows up in the small decisions that compound. When the named engineer’s reputation is attached to the system, they push back on shortcuts the agency’s project manager would otherwise approve. They write the eval that would have been skipped under deadline pressure, because the eval has their name on it. They refuse the prompt that would have leaked PII, because the PII would be in their commit history. The named-individual structure transfers a portion of the agency’s reputational equity to the project itself, and the equity does work the agency’s project manager cannot do.
Onboarding accelerates because the same three people are in most conversation from week 1. The buyer’s engineering team, which will inherit the system, builds direct technical relationships with three named humans rather than trying to keep up with a rotating cast. By week 8, the buyer’s ML engineer and the named retrieval engineer share enough context that the handoff in week 12 takes a day rather than two weeks. The math on this is straightforward: a 12-week engagement with named engineers ships a system that is operable on day 85; the same engagement with rotating staff ships a system the buyer is still onboarding to in month 5. For more on the handoff mechanics, see the anatomy of an AI agency engagement piece, which covers what the first two weeks should produce.
What a good clause looks like
The contract language is shorter than the discussion implies. A working version reads: “Engineer Named A, Engineer Named B, and Engineer Named C are the assigned engineers for this engagement, with allocations of X%, Y%, and Z% of their working time respectively, for the contract duration. Substitution of any named engineer requires (a) at least 14 calendar days written notice to the client, (b) identification of the proposed replacement with a GitHub profile or equivalent portfolio and a list of three prior projects relevant to this engagement, (c) a one-week shadow period during which both engineers attend many standups, design reviews, and code reviews and contribute to PRs, billed at half-rate for the outgoing engineer, and (d) the right of the client to reject the substitution and terminate the affected workstream at no further fee, with delivered artifacts retained. Exceptions for illness exceeding three days, voluntary departure from the agency, and termination for cause are handled under Section X below.”
Section X covers the narrow exceptions. Illness; the agency notifies as soon as the engineer is unable to work three or more consecutive days, and a temporary replacement is approved by the client without the 14-day clock. Departure; when the named engineer leaves the agency, the agency triggers the substitution process immediately with the 14-day clock and absorbs the shadow week as a single billing-line credit, not a separate charge. Termination for cause; if the named engineer is terminated by the agency for misconduct affecting the engagement, the substitution is handled by the agency at no cost to the client and the engagement timeline is extended by the shadow period.
Two structural clauses make the rest hold. A capacity floor: if any named engineer’s allocation drops below 70% of the contracted percentage in any two-week period, the client may renegotiate the project price or schedule. And a transparency clause: the agency provides a weekly time-allocation report showing actual hours per named engineer, signed by each engineer’s manager, surfacing capacity drift before it becomes a delivery problem. For the broader signal landscape on which this sits, the AI agency trust ladder piece covers named-engineer proposals as one of the six markers that distinguish operators from resellers.
When substitution should be acceptable
There are three legitimate substitution events, and the clause should accommodate many three without making them painless. Illness exceeding three days is the simplest: the agency notifies, proposes a temporary or permanent replacement with the same artifacts as the original engineer, and the client approves or rejects within 24 hours. The shadow period is compressed to two days for temporary replacements, full week for permanent.
Voluntary departure is the hardest because the agency has the strongest incentive to manage the news. The clause works only if the agency notifies the client within 48 hours of the resignation, before the departing engineer leaves the agency, while the engineer is still available for a real shadow week. An agency that sits on a resignation for three weeks and then announces a substitution is in breach, and the clause should make that breach contractual.
Termination for cause is rare and the agency absorbs the cost. The client should be notified of the underlying issue at the level appropriate to the engagement; a security violation that affected the project’s data is a different conversation than a generic policy issue. The agency offers a replacement, runs a shadow week at no charge, and extends the timeline to compensate.
What does not count as a legitimate substitution: an internal staffing decision the agency made for utilization reasons, a sales-driven reassignment where a named engineer is moved to a higher-margin account, or a “promotion” that pulls the engineer out of IC work. These are the cases the clause is specifically designed to make uncomfortable for the agency, because they are the cases that destroy buyer outcomes.
How to ask for it without breaking the deal
Buyers worry that demanding named engineers will scare off the best agencies. The opposite is closer to true. Operators; agencies that are confident in their staffing and proud of their engineers; welcome the request and price the rate up by 10–20%. Resellers and body shops resist, because the clause exposes their economics. The request itself becomes a signal in the proposal evaluation: how the agency responds to “we will need named engineers with GitHub profiles and a non-substitution clause” tells you most of what you need to know in the first 30 minutes of the relationship.
The framing that works is collaborative rather than adversarial. The buyer says: we are going to operate this system after you roll off, so the engineers who build it need to be the engineers we get to know, and we are willing to pay for that continuity. The agency that hears this as a partnership move responds with names, GitHub profiles, and a workshop on which workstreams need which engineers. The agency that hears it as a procurement gotcha responds with paragraphs about flexibility and best-fit staffing. The latter response is its own answer.
The buyer who lands the named-individual clause has, in effect, converted a service contract into something closer to a temporary in-house team with a defined transition. The engineering culture of the agency now has a small but real footprint inside the buyer’s organization, and the muscle memory of the system survives the rolloff. That is the outcome AI engagements have been promising for three years and rarely delivering. The proposal is where the promise becomes contractual or stays decorative.
Arthur Wandzel is the founder of SFAI Labs, a forward-deployed AI development agency in San Francisco. He has written more than 60 AI engagement contracts and tracked the named-engineer clause across the post-deploy retrospectives of most one.
Frequently Asked Questions
Why should a buyer require named individual engineers in an AI agency proposal?
Because role-based proposals describe a shape; ‘3 senior engineers, 1 ML lead’; that the agency reserves the right to staff with whoever is available. AI engineering depends heavily on the specific muscle memory of the specific human at the keyboard: the engineer who wrote the retrieval layer is the one who can debug it at 11 p.m. On launch eve. Naming engineers gives the buyer continuity, leverage on small decisions, faster onboarding, and a real handoff to the in-house team that will operate the system after the agency rolls off. Without names, the buyer is paying for a configuration of seats rather than for outcomes.
What artifacts should accompany each named engineer in an AI proposal?
Three artifacts per engineer. A GitHub or equivalent profile; public if available, a curated portfolio of NDA repositories if not. A short list of prior PRs or shipped projects with links and a one-paragraph description of what the engineer specifically owned. A percentage allocation expressed as a number, with a defined floor under which the engagement triggers a renegotiation clause. The proposal should also identify a primary engineer per workstream; typically three names on a 12-week LLM build covering retrieval, agentic tool use, and evals; and each named engineer should have a phone number on the runbook and authorship on the PRs that will be defended in postmortems.
What does a non-substitution clause for AI engineers look like?
Substitution requires (a) at least 14 calendar days written notice to the client, (b) identification of the proposed replacement with the same artifacts as the original engineer including GitHub profile and three relevant prior projects, (c) a one-week shadow period where both engineers attend standups, design reviews, and code reviews, billed at half-rate for the outgoing engineer, and (d) the right of the client to reject the substitution and terminate the affected workstream at no further fee, with delivered artifacts retained. Narrow exceptions are handled separately for illness exceeding three days, voluntary departure, and termination for cause.
What does the agency lose by naming engineers in a proposal?
Less than agencies typically fear. Lock-in is real but small; the agency that names three engineers on one project still has flexibility on the other twenty. Utilization is recoverable through higher rates, since buyers pay more for specific people than for fungible bodies. Vacation is a paperwork problem solved by scheduling shadow weeks at the front of the project. The mid-project departure is the genuinely hard case and is what the substitution clause’s ‘departure’ exception is for. What the agency loses is the ability to bait-and-switch; pitching the founder and delivering the junior, or a ‘team of 3’ that turns out to be one staff engineer split across four projects. Reputable agencies stopped doing those years ago and will not feel the loss.
Beyond continuity, what does the buyer gain from named-engineer proposals?
Two compounding wins beyond continuity. Leverage: when the named engineer’s reputation is attached to the system, they push back on shortcuts the agency’s project manager would otherwise approve, write the eval that would have been skipped under deadline pressure, and refuse the prompt that would have leaked PII. The named-individual structure transfers a portion of the agency’s reputational equity to the project itself. Onboarding speed: the buyer’s engineering team builds direct technical relationships with three named humans rather than tracking a rotating cast, so the handoff in week 12 takes a day rather than two weeks, and the system is operable on day 85 instead of in month 5.
When is engineer substitution legitimately acceptable mid-project?
Three legitimate substitution events. Illness exceeding three consecutive days; agency notifies, proposes a replacement with the same artifacts, client approves within 24 hours, shadow period compressed to two days for temporary or full week for permanent. Voluntary departure; agency notifies the client within 48 hours of resignation, before the engineer leaves, while they are still available for a real shadow week. Termination for cause; agency absorbs the cost, runs a shadow week at no charge, and extends the timeline. What does not count: internal utilization decisions, sales-driven reassignments to higher-margin accounts, or ‘promotions’ that pull the engineer out of IC work.
Will demanding named engineers and a non-substitution clause scare off the best AI agencies?
The opposite is closer to true. Operators; agencies confident in their staffing and proud of their engineers; welcome the request and price the rate up 10-20%. Resellers and body shops resist because the clause exposes their economics. How an agency responds to ‘we will need named engineers with GitHub profiles and a non-substitution clause’ is itself a high-signal data point in the proposal evaluation. The framing that works is collaborative: ‘we will operate this system after you roll off, so the engineers who build it need to be the engineers we get to know, and we are willing to pay for that continuity.’ Operators respond with names; resellers respond with paragraphs about flexibility.
What is a capacity floor and why does it matter alongside named engineers?
A capacity floor is a contractual minimum for how much of each named engineer’s time is dedicated to the engagement. A working version reads: ‘if any named engineer’s allocation drops below 70% of the contracted percentage in any two-week period, the client may renegotiate the project price or schedule.’ It matters because naming engineers without a capacity floor permits the soft bait-and-switch where the engineer is technically still on the project but is splitting attention across three other accounts. Combined with a weekly time-allocation report signed by each engineer’s manager, the capacity floor surfaces drift before it becomes a delivery problem.
Does the named-engineer clause work for short engagements or only for multi-month builds?
It works for both, but the structure differs. On a short engagement of four to six weeks, naming a single primary engineer per workstream is sufficient and the shadow period for substitution can be compressed to two days. On a multi-month build of twelve weeks or more, the proposal should name three engineers, identify the primary engineer per workstream, and require the full one-week shadow period. The shorter the engagement, the higher the cost of any substitution as a fraction of total schedule, which counterintuitively makes the clause more important rather than less for short builds.
How does naming engineers interact with the agency’s IP and confidentiality obligations?
It does not weaken them. The agency still owns the work product structure defined elsewhere in the contract, the engineers are still bound by their employment agreements with the agency, and confidentiality flows from the master services agreement rather than from the staffing clause. What changes is transparency about who is doing the work; a name and a portfolio link, not a disclosure of proprietary methods. For engineers under NDA on prior work, the proposal can describe the project at the level of ‘retrieval system at scale X for sector Y’ rather than naming the client. The named-engineer clause coexists cleanly with standard agency IP and confidentiality terms.
Arthur Wandzel
