The invoice is not the whole price.

Video caption

The invoice is not the whole price.

Include work that never appears in the token meter.

Custom code can be justified by a capability the market does not offer.

My rule: Review the decision after real operating evidence arrives.

Narration uses Eric's authorized AI voice clone.

#EricFieldNotes

Full written post / accessibility read

Suppose a small company replaces a maintained approval tool with an agent-built service. The code-generation bill is small and the demo is convincing. Neither figure tells us what it costs to operate the service through a policy change, an employee departure and a customer dispute.

List license expense for the purchased option. For the custom option list the named owner, engineering review, security maintenance, integration changes, backups, support, data export, incident recovery and the business hours diverted from the team's primary work. Assign an interval, not a fabricated exact cost.

A unique approval policy that wins customers may justify the commitment. A replacement scheduler with the same features as a maintained product probably does not. Compare accepted business outcomes and customer risk under both routes, not the size of the first pull request.

Name the product owner, acceptance cases, recovery path, export format and a date to revisit cost and reliability. If differentiation never appears or maintenance consumes the promised gain, migrate back. Do this because build versus buy is an ongoing operating decision, not a one-day verdict from a demo.

Narration uses Eric's authorized AI voice clone.

#EricFieldNotes

Four-beat scene transcript

1. The invoice is not the whole price.

Suppose a small company replaces a maintained approval tool with an agent-built service. The code-generation bill is small and the demo is convincing. Neither figure tells us what it costs to operate the service through a policy change, an employee departure and a customer dispute.

Visual: A cheap first build can create an expensive product commitment.

2. Write the recurring ledger.

List license expense for the purchased option. For the custom option list the named owner, engineering review, security maintenance, integration changes, backups, support, data export, incident recovery and the business hours diverted from the team's primary work. Assign an interval, not a fabricated exact cost.

Visual: Include work that never appears in the token meter.

3. Separate commodity from advantage.

A unique approval policy that wins customers may justify the commitment. A replacement scheduler with the same features as a maintained product probably does not. Compare accepted business outcomes and customer risk under both routes, not the size of the first pull request.

Visual: Custom code can be justified by a capability the market does not offer.

4. Add a stop criterion.

Name the product owner, acceptance cases, recovery path, export format and a date to revisit cost and reliability. If differentiation never appears or maintenance consumes the promised gain, migrate back. Do this because build versus buy is an ongoing operating decision, not a one-day verdict from a demo.

Visual: Review the decision after real operating evidence arrives.

Research and claim limits

• DORA 2025 State of AI-assisted Software Development (S225)
• GitHub Copilot Agents application card (S226)
• NIST AI RMF Core (S227)

The examples identified as illustrative or simulated are design probes, not reported incidents. Vendor specifications do not establish workload performance.