Appendix A. Cost Economics

How much does agentic coding cost?¶

The second most common question from engineering managers after "does it work" is "what does it cost." Specific prices change every quarter; the rubric does not. Plug your team's numbers into the structure below.

Pricing tiers

Three pricing models dominate, in roughly this order of complexity:

Per-seat. Flat fee per engineer per month, regardless of usage. Predictable. Cheap for occasional users, expensive for power users who run the agent all day.

Per-token (metered). You pay for input tokens the agent reads and output tokens it produces, summed across sessions. Fair to occasional users; punishing for the engineer running three agent sessions in parallel for ten hours.

Enterprise. Negotiated annual commitment bundling seats plus a token allowance plus the compliance surface (Zero Data Retention, audit log access, dedicated capacity, SSO, signed BAAs). Amortizes both extremes and adds what regulated industries require.

The bounding heuristic

Cost-per-engineer-per-month for an agentic coding tool is bounded by what your team would otherwise have spent on similar tooling: IDE licenses, code-intelligence platforms, AI-assisted-coding subscriptions, plus a fraction of a senior engineer's hourly time saved per week. If the agent's all-in monthly cost-per-engineer exceeds that envelope by a wide margin, the math is unlikely to work regardless of vendor. If it falls comfortably inside, the math is unlikely to fail.

Run the calculation for your own team. Take your current tooling stack, your engineers' loaded hourly cost, an honest estimate of how many hours per week the agent will save them, and the agent vendor's per-seat or per-token quote. The break-even point lands quickly when the agent saves even a few hours per month per engineer.

What is not in the sticker price

The vendor's quote is the easy part. Four categories are not in it and dominate the real total cost of ownership.

Integration. Writing your custom skills, configuring your hooks, setting up MCP servers for internal systems. One-time investment of engineer-weeks; pays back over the lifetime of agent use.

Skill-authoring time. Maintaining AGENTS.md, writing and updating skills as the codebase evolves. Ongoing; typically a few hours per engineer per month, plus concentrated time from the team's champion (Chapter 10).

Review time. Reviewing agent output. Less per change than reviewing hand-written code in most cases, but not zero, and concentrated on senior reviewers.

Governance overhead. Security review through your CISO. Zero Data Retention addendum negotiation. Procurement cycle time. Audit logging infrastructure. Vendor-risk monitoring. Variable by company; ranges from a week to a quarter.

A worked example¶

Chapter 10's manager sidebar left a 20-engineer financial-services team mid-arc: 41% of merged PRs agent-touched in month two, cycle time on that set 28% below the pre-agent baseline, defects within noise. Run that same team through this appendix's rubric. The cost figures below are round numbers for the arithmetic, not quotes - plug in your own; the appendix's whole point is that the specific ones go stale by next quarter.

Start with the sticker, the easy part. The engagement put 13 engineers on the Team tier and 7 on Pro seats, because those 7 use the agent rarely and the tier matched the usage (Chapter 10's point: bounded spend, not uniform tooling). At an illustrative $30 per Team seat and $20 per Pro seat, that is 13 seats at $30 plus 7 at $20, or $530 a month, call it $6,400 a year. Write that number down. It is the smallest one on the page.

Now the four categories that are not in the sticker, at the magnitudes the book already uses:

Price the human lines at the same honesty. A few engineer-weeks of integration, at any realistic loaded cost, clears the first year's sticker on its own, once. Skill-authoring - a few hours per engineer per month across 20 engineers, plus the champion's part-time quarter - is a recurring line that can run larger than the seats themselves. Review time lands on the senior reviewers, your most expensive hours. Governance in a regulated shop is a week to a quarter of security, procurement, and audit-logging work before a single seat is billed. The sticker is real, but it is not where the money is.

The value side uses the bounding heuristic in reverse. Take an engineer at an illustrative $100 loaded per hour. A $30 Team seat is covered when the agent saves that engineer roughly twenty minutes across the whole month; a few saved hours a month is not close. The seat clears its own bar early and easily. But that back-of-envelope is not what justified the spend. What did was the operational number the manager already had: 41% of merged PRs agent-touched, cycle time on them 28% below baseline, defect rate flat. Measured hours out, not projected dollars.

That is the discipline. The sticker is the smallest line and the one everybody asks about first. The four TCO categories dominate the real total, and the human ones dwarf all the seat math. And the number the manager defends to the board is the operational one - 28% lower cycle time at no measurable change in defects - not a projected ROI the dashboard cannot yet support (Chapter 10). Pricing changes; this arithmetic does not.

Pricing changes; the math does not

Specific prices in any quarter will be wrong the next quarter. The shape of the math will not. Per-seat scales with team size; per-token scales with usage intensity; enterprise plans bundle both with compliance. The bounding heuristic and the four-category TCO list survive every pricing change. Walk into the procurement conversation in Chapter 10's manager section with your own numbers in this rubric.