There is a massive flaw in how most engineering organizations treat artificial intelligence: engineering leaders can't isolate AI's contribution because it's baked into baseline velocity.
The prevailing advice for enterprise AI adoption is straightforward: treat an AI agent like a new hire. In the engineering space, this isn't just a convenient metaphor. It's a pretty accurate financial model. When you deploy a coding assistant or an autonomous agent, you are making a direct trade-off. You are exchanging software licenses and API costs for the equivalent output of a developer writing boilerplate, mapping dependencies, or generating tests. Comparing AI spend directly to headcount makes perfect sense on a balance sheet. It is a predictable, operational expense.
But when AI is just a baseline tooling cost, its financial impact gets blended into the general "velocity" of the engineering organization. Sure, if you have the right structures and details in your ticket tracking tool, you could get pretty close to inferring how much AI you used with it, but a lot of people don't have that, and a lot of those that do get a ballpark vs something they can clearly point to. It becomes very hard to prove to finance exactly what that monthly subscription is buying. As a result, all you can do is view its impact based on the overall delivery of the engineering department. You can't specifically pinpoint the benefit and say "AI did this for us". At best, you can say that as an organization you delivered faster than you have historically.
To get the most out of these tools, and to solve the ROI tracking problem, engineering leadership needs to view AI not just as a permanent baseline "virtual employee," but as the ultimate temporary contractor.
The Baseline vs. The Burst (The New Staff Augmentation)
When you hire an engineer, or provision an AI agent to do an engineer's job, you are permanently expanding your baseline capacity. That covers your day-to-day velocity.
But what happens when you face a massive, time-boxed initiative? Consider the strategic heavy lifting of breaking down a complex legacy monolith into distributed, cloud-native systems. There are intensive phases of this work that require a temporary spike in effort.
In traditional resourcing, when you hit these spikes, you bring in temporary contractors or consultants. You pay a premium for short-term velocity.
Project-based AI spend is the modern equivalent of staff augmentation, but without the friction. If you view AI strictly as baseline headcount, your only option is to "hire" more AI agents permanently, which inflates your annual budget. However, if you treat AI as a project-specific line item, just like a six-month contractor budget, you can "burst" your capabilities for the duration of that workstream.
Even better, "AI contractors" come without the traditional headaches:
- Zero Ramp-Up Time: Human contractors take weeks to understand your architecture and domain logic. An AI agent powered by your repository's context is productive on day one.
- No Agency Fees: You are paying strictly for compute and API calls, not overhead.
- Instant Offboarding: When the project ends, the spend spins down immediately.
Making the Math Crystal Clear with "Good, Better, Best"
When you apply this contractor mindset and attach a specific AI budget to a time-boxed project, the ROI math becomes crystal clear. You are separating the cost from your general operating budget.
Now, assuming you have clear, definitive success metrics for all of your initiatives, this lets you tie your AI spend directly to the performance of those success metrics. You are no longer guessing at broad productivity gains; you are tying an exact dollar amount directly to the measurable business outcome for a specific initiative. You can definitively say, "Spending $5,000 on AI accelerated this specific project"
The most effective way to establish these clear financial expectations upfront is by baking this contractor mindset directly into a "Good, Better, Best" proposal. "Good, Better, Best" is a simple decision-framework used to present leadership with tiered options rather than a single yes/no ask. Each tier trades a different lever (time, cost, scope) for a different outcome, so the business can choose its risk tolerance instead of being boxed into one plan. I use this format regularly to bridge technical execution and boardroom strategy: it lets leaders weigh the pros and cons of a decision without needing the full technical picture underneath it.
You can pull a number of levers to move from Good to Better or Best. Staff augmentation, scope, security posture, and so on. There's no universal right answer. It depends on the goals of the initiative and the constraints of the product. For this example, the goal is compressing the timeline. Historically I've pulled that lever through scope cuts or staff augmentation. Here, I'll pull it with AI spend instead.
Good | Better | Best | |
Timeline | 18 Months | 15 Months | 12 Months |
Workstreams | 3 Concurrent | 3 Concurrent | 3 Concurrent |
Teams | 4 Teams, ~20 FTE | 4 Teams, ~20 FTE | 4 Teams, ~20 FTE |
AI Ask | 0 | 1000/mo per FTE | 2000/mo per FTE |
Team Cost | ~4.5M | ~3.75M | ~3M |
Additional AI Ask | 0 | ~300K | ~480K |
Total Cost | ~4.5M | ~4M | ~3.5M |
Additional Budget Ask | 0 | ~300K | ~480K |
The Boardroom Pitch
Framing AI this way shifts the conversation from abstract productivity metrics to concrete project ROI. You acknowledge the reality that baseline AI is essentially headcount, but you offer the business a new, efficient way to handle the project spikes.
You can clearly state, "If we want to hit the 'Best' tier for this effort, we don't need to sign a 9–12 month contract with an external consultancy locking us into a year of 5–10 developers for $1M–$2M dollars. Instead, we can temporarily increase our AI seat license spend with a one-time capped cost of $480K in dedicated AI compute specifically allocated for this workstream."
I've run this exact exercise on projects where the only lever available was scope or headcount, and it always meant a harder conversation. I've always had to either cut something the business wanted, or sign a contract that outlasted the actual need. AI spend is the first lever I've had that scales down as cleanly as it scales up.
Next time your team is facing a compression ask, don't reach for the contractor budget first. Save that for when you need specialized skills or knowledge. Instead, build the Good/Better/Best, put AI spend in the "Better" and "Best" columns, and let the business choose its own risk tolerance with a number attached to every tier.