About the author: I'm Charles Sieg, a cloud architect and platform engineer who builds apps, services, and infrastructure for Fortune 1000 clients through Vantalect. If your organization is rethinking its software strategy in the age of AI-assisted engineering, let's talk.
Seven tasks. April 10 was the lowest-volume day of the three-day stretch, but the character of the work was qualitatively different from the prior day's phase-by-phase builds. The session covered a large-scale website migration (five sites, 84 pages), a security architecture change removing mutual TLS from the tools fleet and replacing it with a WAF, accessibility finalization across 22 React repos, four separate AI-driven product innovations, a static site build for a legacy domain, a round of issue tracker fixes, and a corporate website HTML repair. The two longest tasks in terms of Claude time (120 minutes and 155 minutes) both involved multi-service, multi-repo coordination with external dependencies.
The weighted average leverage factor was 112.7x with a supervisory leverage of 1,352.9x. Total human-equivalent work was 298 hours across 7 tasks.
Task Log
| # | Task | Human Est. | Claude | Sup. | Factor | Sup. Factor |
|---|---|---|---|---|---|---|
| 1 | Four radical innovations: list template tool gallery (30 templates, 6 categories); static site generator auto-linking via semantic embeddings (46 tests); accounting tool LLM auto-categorization for bank transactions; newsletter tool AI A/B subject line testing (toggleable, full lifecycle) | 120h | 32m | 3m | 225.0x | 2400.0x |
| 2 | Accessibility final completion: all 22 React repos passing axe tests with 49 total tests + 20 per-activity tests + 6 real WCAG violations discovered and fixed | 40h | 40m | 2m | 60.0x | 1200.0x |
| 3 | Issue tracker fix: create a legacy domain static site via static site generator - 786 files, 8 content pages, full template set, 720 domain data files, SCSS design system | 16h | 12m | 1m | 80.0x | 960.0x |
| 4 | Repair corrupted HTML on 14 corporate website tool pages: restore feature cards from git history, fix section nesting, move accessibility groups, build and deploy | 6h | 10m | 3m | 36.0x | 120.0x |
| 5 | Migrate 5 websites to static site generator (list template, credential system, enterprise, corporate, product site - 84 pages including React SPA teardown) | 65h | 155m | 6m | 25.2x | 650.0x |
| 6 | Issue tracker fixes: triage board replaces bug reports board, sidebar state persistence, project rename/archive/delete UI, create-dedup in reducer, card detail modal loading fallback, vitest-axe type augmentation for CI | 11h | 29m | 3m | 22.8x | 220.0x |
| 7 | Remove mTLS from tools fleet: disable ALB mTLS, add WAF (3 rules), strip MCP configs, fix static site generator crash (missing wheel), fix accounting tool diagnostics + config + DB, remove probeUrl from 5 frontends, fix newsletter tool JWT kid matching, fix newsletter tool port conflict, fix 3 frontend TypeScript type errors | 40h | 120m | 15m | 20.0x | 160.0x |
Aggregate Statistics
| Metric | Value |
|---|---|
| Total tasks | 7 |
| Total human-equivalent hours | 298 |
| Total Claude minutes | 398 |
| Total supervisory minutes | 33 |
| Total tokens | 3,066,000 |
| Weighted average leverage factor | 112.7x |
| Weighted average supervisory leverage factor | 1,352.9x |
Analysis
The four-innovations task (225.0x) was the standout. Four separate AI-native features were designed and implemented in 32 minutes across four different codebases: a template gallery with 30 curated items across 6 categories for the list template tool; an auto-linking engine for the static site generator that uses semantic embeddings to detect and inject cross-page links (with 46 tests); an LLM-based transaction auto-categorization system for the accounting tool; and a full A/B subject line testing feature for the newsletter tool including a lifecycle manager, toggleable activation, and persistent result tracking. Each of those represents a complete feature, not a prototype. The 120 human-hour estimate reflects what a product engineering team would budget across those four independent efforts.
The mTLS removal (20.0x) and the website migration (25.2x) were the day's lowest-leverage tasks, and both for the same reason: both required navigating real infrastructure state rather than generating code from a specification. Removing mTLS from a live fleet means verifying each service's behavior at each step (disable, redeploy, test connectivity, verify WAF rules, check dependent services). The website migration ran 155 minutes because 84 pages across 5 sites with varying template structures, asset pipelines, and deployment configurations required iterative attention rather than bulk generation.
The accessibility finalization task (60.0x) closed the multi-day WCAG 2.1 AA remediation thread that started on April 9. At 40 minutes of Claude time and 22 repos in scope, it produced a complete pass: 49 axe test files, 20 per-activity coverage tests, and 6 genuine WCAG violations found and fixed in production components. The supervisory leverage on this task was 1,200x because the entire audit-and-fix sequence was initiated with a two-minute prompt referencing the prior day's tier-2 remediation output.
At 7 tasks and a weighted leverage of 112.7x, April 10 was the lowest-output day of the three. The infrastructure-heavy character of the session is directly reflected in the leverage numbers: two of the three lowest factors belong to tasks where iterative external state validation was unavoidable. The supervisory leverage of 1,352.9x is the highest of the three days because several long-running tasks were initiated with short, high-trust prompts referencing prior session context rather than detailed new specifications.
Let's Build Something!
I help teams ship cloud infrastructure that actually works at scale. Whether you're modernizing a legacy platform, designing a multi-region architecture from scratch, or figuring out how AI fits into your engineering workflow, I've seen your problem before. Let me help.
Currently taking on select consulting engagements through Vantalect.