Opening Hook

I claimed “2 weeks of work in 2.5 hours” yesterday. Someone called BS. They were right. Here are the actual numbers — and why 3X is still impressive.

The Real Numbers

Task Manual Time AI-Assisted Time Speedup
GitHub profile patch (clone + edit) 15 min 15 sec (automated) ~60×
Resume rewrite (3 bullets) 60 min 5 min (generate + review) ~12×
Site build (207 posts) 30 sec 30 sec 1×
Write 3 blog posts (~1,500 words) 4–6 hrs 15 min gen + 5 min review each ~12–20× generation/review rate
Install + configure OPA/Rego 30 min reading docs 10 min + trial/error ~3×
PDPA scans + deploys 15 min per cycle 5 min automated ~3×
Careerbot fixes (autofill + LLM chain) 8–12 hrs debugging 2 hrs guided ~5×

Estimated manual total: ~6–8 hours of concentrated work
Observed wall-clock session: ~2.5 hours
Session-level result: roughly 2.4–3.2× against that estimate

The table contains task-level rates; the session claim is an estimate, not a controlled experiment.

Why the “2 Weeks” Claim Was Wrong

Marketing exaggeration. Let me decompose what happened:

  1. Waiting time isn’t productive time — builds, deploys, model inference all paused the clock
  2. Context switching — moving between blog writing, resume edits, tooling — killed flow
  3. Review overhead — every AI-generated output required human verification (rightfully so)

The real win wasn’t compression. It was focus.

Getting from idea → production in one 2.5-hour session is worth more than 2 weeks of scattered context-switching.

Honest Measurement Framework

Diagram

How to Measure Honestly Going Forward

Track these separately:

  1. t_generate — time AI spends producing artifacts
  2. t_review — time you spend verifying/editing output
  3. t_wait — builds, deploys, model inference (not productive work)
  4. t_context — meetings, emails, interruptions

True productivity multiplier:

t_manual_equivalent / (t_generate + t_review + t_wait + t_context)

In yesterday’s session: - Total wall clock: 2.5 hours - Productive work: ~1.5 hours (t_generate + t_review) - Wait time: ~1 hour (builds, deploys, model calls) - Context switching: minimal

So the session multiplier is roughly 2.4–3.2×, assuming the 6–8 hour manual estimate is comparable.

Why This Matters

Overstated productivity claims create unrealistic expectations. Understated ones breed complacency.

3× sustained velocity is still worth celebrating: - Can turn some 1-week tasks into shorter iteration cycles - Enables same-day iteration cycles - Frees time for strategic work (the actual value creation)

Action Items

  1. Time-box AI sessions — track generate vs review time separately
  2. Batch I/O-bound tasks — queue builds during model inference windows
  3. Measure end-to-end session velocity — not just artifact generation speed
  4. Drop the “2 weeks” framing — it’s misleading and unsustainable

Takeaway

Honesty about velocity builds sustainable workflows. 3× is already transformative — no need to inflate it to 10×.

The real productivity gain isn’t raw speed. It’s unbroken focus loops that would normally be shattered by meetings, builds, and context switches.