What Is This Note?
This is a dated note on our February 2026 case study, "We Built a Production AI Travel Platform in 1 Day. Here Is How." Lakbay AI is real, and a demo is live. The headline claims about how fast and how safely we built it are not supported by our records, so this note keeps the product facts and withdraws the rest.
What Is Lakbay AI?
Lakbay AI is an AI travel concierge for the Philippines. A traveler describes a trip in conversation and gets a structured, day-by-day route. It retrieves structured travel content across 18 mapped destinations, searches travel options through the Amadeus API, and has three connected portals for travelers, agents and admins. A demo is live at travel-conceirge.vercel.app, and our project page has the details.
What Does the Repository History Show?
Our private repository shows the build happened over several days, with most of the early commits landing on 1 and 2 February 2026, and work continuing for months after the case study.
Lakbay AI commits per day, up to the case study
Git history, dates in UTC
Two busy days of building is a fast start, and we are glad to say so. It is not "built in 1 day", and a project that keeps receiving commits for seven months was not finished on launch day.
Which Claims From the Original Post Are Withdrawn?
| In the February 2026 case study | What we can show | Status |
|---|---|---|
| Built in 1 day | 35 commits across five days by the case study date; most early work on 1 and 2 February | Withdrawn |
| A 70x speed-up over an agency and 130x over a solo developer | Compared with other companies’ quoted ranges, not a measured project | Withdrawn |
| Zero critical vulnerabilities at launch | No saved scan result from launch | Withdrawn |
| A near-zero cost compared with agency budgets | No cost record kept for the build | Withdrawn |
| CLAUDE.md described as enforcing security-first patterns | CLAUDE.md instructs the agent; enforcement comes from tools and database policies | Corrected |
| Proof that Z-threads compress timelines by a set multiple | One project with an overstated timeline proves no multiple | Withdrawn |
What the case study said, and what we can show
February 2026 post versus our records in October 2026
- Built in 1 day
- A 70x speed-up over an agency
- Zero critical vulnerabilities
- 35 commits over five days, then months of work
- No measured comparison
- No saved scan from launch
What Holds Up?
The product holds up: Lakbay AI exists and the demo is live. The working method holds up as a method, not as a multiplier. Thread-Based Engineering puts a person at the start and the end of every thread and lets automated gates do the rest; Part 2 of this series describes the gates we actually run, with counts from our own repository.
Lakbay AI, in verified numbers
Project page and repository, checked 2 October 2026
mapped destinations
connected portals
commits by the case study date
claims withdrawn or corrected
What This Series Covers
This is Part 4 of the AI Technical Debt series, now a dated note. Part 1 covers the data on AI-generated code, Part 2 covers how we govern AI-assisted development, and Part 3 covers practical mitigation patterns.
Lakbay AI: Questions Readers Ask
Common questions about this topic, answered.
What is Lakbay AI?
Lakbay AI is an AI travel concierge for the Philippines that we built. It plans trips from a conversation, retrieves structured travel content across 18 mapped destinations, searches travel options through the Amadeus API, and has three connected portals for travelers, agents and admins. A demo is live at travel-conceirge.vercel.app.
Was Lakbay AI built in one day?
No. Our private repository shows a first commit on 30 January 2026, 21 commits on 1 February, 9 on 2 February and 35 by the date of the case study, 8 February, with work continuing for months after. Most of the early work landed over two days, but "built in 1 day" overstated it, so we withdrew it.
Was it 70 times faster than an agency?
We cannot show that. The 70x figure compared our build time with agency price-and-timeline ranges quoted by other companies, not with a measured project of the same scope. With the one-day figure withdrawn as well, the comparison has nothing to stand on.
Did Lakbay AI have zero critical vulnerabilities at launch?
We have no saved scan result from launch, so we withdrew the claim. A claim of zero vulnerabilities needs a dated scan report behind it.
Does a CLAUDE.md file enforce security rules?
No. A CLAUDE.md file is instructions to the coding agent, and an agent can miss an instruction. Enforcement comes from tools that run regardless: database security policies, lint rules, type checks, tests and required checks before merge. The case study described CLAUDE.md as enforcement, which it is not.
Which technology does Lakbay AI use?
Next.js 16 and TypeScript for the app, OpenAI for generation, Supabase pgvector for retrieval over the destination content, and the Amadeus API for travel search, as listed on our Lakbay AI project page.
What is a Z-thread?
In Thread-Based Engineering, a Z-thread is the most autonomous thread type, where the human checkpoint moves to the end only. We reserve it for well-tested, low-risk work. The case study presented Lakbay AI as proof that Z-threads compress timelines by a set multiple; that claim is withdrawn.
What changed in this post on 2 October 2026?
We cut it down to a dated note. The one-day timeline, the 70x and 130x speed figures, zero critical vulnerabilities at launch, a near-zero cost comparison and the description of CLAUDE.md as enforcement are withdrawn. What Lakbay AI is, and what its repository history shows, are kept.
