A CFO I know once slid two PDFs across a table and asked me to explain the world to her. On the left, an agency quote for a new operations platform: $85,000, fixed scope, delivery in four months. On the right, a proposal for an embedded engineer: $22,000 a month, open-ended, first release in three weeks. "The second one costs more," she said, "but everyone keeps telling me it's cheaper. Explain."

Both documents were telling the truth, and both were lying, which is what makes the software agency vs embedded engineer cost comparison so reliably confusing. The agency PDF was honest about the build cost and silent about everything around it. The FDE proposal was honest about the monthly burn and silent about how few months it would actually run. The CFO was holding a quote and a rate, and neither one was a cost.

The honest answer comes down to this: agencies quote lower and almost always deliver slower, and the gap between the quote and the invoice is filled with your own management hours, change orders, rework, and the cost of waiting. An embedded engineer looks expensive per month and cheap per outcome. Run the total-cost math before deciding which PDF to sign.

The quote vs. the invoice

Fixed-price agency quotes have a structural secret: the fixed price is fixed to a scope document, and the scope document was written before anyone understood the problem. The first time reality disagrees with the document, which happens around week three, a change order appears. Illustrative but painfully typical: the $85,000 quote becomes $113,000 in change orders, plus a $27,000 "phase two" for the things that were always needed but somehow didn't make the original scope.

None of this is fraud. It's the business model working as designed, and the agency would argue, fairly, that the client kept changing its mind. Of course the client changed its mind. Learning what you actually need is the entire point of building software, and a contract that charges you every time you learn something is a contract that taxes progress.

The hidden line items

The invoice gap is only part of it. The rest lives on your side of the ledger, in costs that never appear in any quote:

The speed multiplier nobody prices

Four months to first release versus three weeks isn't a scheduling detail; it's a financial instrument. Say the platform is meant to replace a manual process that burns $9,000 a month in labor and errors, which is a modest figure for a mid-size operation. The agency path burns roughly three more months of that process before v1 arrives, call it $27,000 in pure waiting cost. The embedded path kills the manual process before the agency would have finished its discovery workshops.

Speed compounds in the other direction too. The embedded engineer ships in week three, which means real user feedback arrives in week four, which means months two through four are spent improving something people use instead of building toward a guess. This is the argument that usually lands with finance, and it's the same one behind selling an FDE engagement to a CFO: you're not buying engineering hours, you're buying a shorter distance to value.

There's also a morale line item nobody models. Four months of waiting teaches your team that the new system is a rumor; three weeks teaches them that change is something that happens to their actual work. The second belief is worth money, because the next improvement request arrives as a conversation instead of a work order.

The total-cost math, with numbers

Assemble the illustrative comparison honestly. Agency path: $85,000 quote, plus $28,000 in change orders, plus $36,000 of internal management time, plus $27,000 of waiting cost, landing near $176,000 with first value at month four and a codebase you now need to staff for. Embedded path: four months at $22,000 all-in, $88,000 total, first value in week three, iteration included in the rate, and the knowledge stays because the engineer sat with your people the whole time.

The embedded path comes out roughly half the total cost in this scenario, and the scenario isn't rigged; if anything, the agency numbers are kind. The full FDE cost breakdown shows where the monthly rate actually goes, and the hidden cost of the manual processes you're replacing is the multiplier that makes waiting so expensive. Quote-to-quote, the agency wins. Cost-to-value, it isn't close.

Play with the assumptions and the conclusion survives. Halve the management hours, forgive the change orders entirely, and the paths land within striking distance of each other on dollars alone; the waiting cost still tips it, and the embedded path still ships three months earlier. The agency quote only wins if you value your own team's time at zero, which is a fun accounting choice until you meet payroll.

When an agency is actually the right call

Honesty cuts both ways, so here's where the left PDF deserves the signature. Marketing sites and brand work: agencies are genuinely better at this and always will be. Fully specified, genuinely fixed projects: if you can write the whole spec today and it won't change, fixed price works as advertised. And the surge-capacity case: you have a strong internal team that needs twenty extra hands for three months, not a brain.

The pattern is that agencies win when learning is not part of the project. The moment the real answer is "we'll know what we need when we see it," you're paying for iteration either way, and you get to choose between an iteration tax and an iteration-inclusive rate. Most operations platforms, internal tools, and automation builds fall squarely in the learning camp, which is why the comparison so often ends the way the CFO's spreadsheet did.

She signed the right PDF, eventually, after adding three rows to her spreadsheet: management hours, change orders, waiting. The agency number grew by 107 percent. The FDE number didn't move at all, which, she noted, was the first time a vendor estimate had ever survived contact with her arithmetic.