TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A report by Dan Luu argues that improved AI-assisted coding lets developers attempt software projects that once seemed impractical. It points to pgrust and several personal experiments as examples, while warning that prototypes still need serious testing and benchmarking before they are ready for wider use.
Dan Luu’s report argues that software developers can now attempt projects that would previously have been too costly to build, as AI-assisted coding reduces the time needed to reach a working prototype. The essay uses pgrust, a Rust-based PostgreSQL project, and Luu’s own experiments to make the case, while emphasizing that faster development does not remove the need for testing or product-quality work.
Luu describes how Michael Malis, who is developing pgrust, has argued that developers should be more ambitious about what they build. Luu says he trusts Malis’s work, but acknowledges that a reader encountering pgrust without that familiarity might dismiss it as another low-quality AI-generated rewrite. The essay recognizes that skepticism as understandable given the volume of projects that make performance claims without strong evidence.
In his account, testing pgrust has also exposed bugs in PostgreSQL, including issues involving numeric operations, JSONPath results, pgcrypto, memory safety in a dictionary loader and money overflow. Luu says those findings suggest substantial testing is taking place, though he does not claim that pgrust itself is bug-free. He argues that identifying defects in established software can produce lasting value even if the new project does not succeed.
Performance is pgrust’s main pitch, according to Luu, but the report does not provide a comprehensive set of independent benchmarks establishing its performance against PostgreSQL or other systems. Luu also describes building a workload-tuned regex engine, a ripgrep fork and a Rust build tool. He says these experiments took a low single-digit number of hours of his time to reach a “works for me” level; he cautions that preparing them for use as products or open-source projects would require substantially more work.
From Prototype to Useful Software
The report’s central point is about what developers choose to attempt, not a claim that AI-generated software is ready to trust without review. If coding tools let individuals test ideas that once demanded much more time, developers may be able to explore specialized tools, performance improvements and alternative implementations that would otherwise remain unbuilt. That could benefit users even when a project does not become a successful replacement: Luu’s example is pgrust testing that reportedly finds defects in PostgreSQL.
There is also a practical risk in treating speed of production as proof of quality. Luu notes that low-quality projects and weak benchmark claims are widespread, making it difficult for readers and potential users to distinguish serious engineering from convincing presentation. His argument depends on a combination of ambition and verification: developers can attempt more, but testing, realistic benchmarks and further engineering remain necessary before a prototype is suitable for broader use.
Products Worth Considering
Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Black
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
As an affiliate, we earn on qualifying purchases.
The Case Behind the Argument
The essay places its argument amid growing skepticism toward software rewrites and projects associated with AI coding tools. Luu says many such projects are dismissed as “vibeslop” because low-quality examples are common and people rarely have time to inspect each new release. He says he initially relied on his trust in Malis rather than investigating pgrust, and later checked the project while considering it as an example for the essay.
Luu connects the opportunity to faster coding models and the bottlenecks they can expose. He says he considered what might slow his computer if he were using a leading model at speeds 10 to 100 times higher, citing search, Rust compilation, system calls and filesystem operations. The report describes his own experiments and observations; it does not present a controlled study showing that AI tools reduce project times by a fixed amount across developers or software categories.
One example is Luu’s account of building tools despite limited prior knowledge of areas such as regex design, string-matching algorithms and compilers. He says AI-assisted work made these personal experiments feasible in hours rather than the year or more he estimates they might previously have taken him to reach a working state. That estimate is his own, and applies to projects at a personal “works for me” standard—not to finished, maintained software.
“We should be more ambitious with the software we build.”
— Michael Malis, as quoted by Dan Luu
What the Examples Do Not Establish
The supplied report does not provide enough benchmark detail to establish pgrust’s performance across workloads, or to compare it independently with PostgreSQL, ClickHouse or other database systems. It also does not establish how many reported PostgreSQL bugs have been confirmed, fixed or released; Luu describes them as bugs found through testing, but the material gives no complete status update for each issue.
The report’s estimates of development time come from Luu’s personal experience, not a controlled comparison. It remains unclear how reliably other developers can reproduce those results, how much supervision the coding tools required, and how the approach performs on larger projects. The essay also acknowledges that moving from a personal prototype to software that is useful and maintainable for others takes considerably more work.
Testing Before Wider Adoption
The next meaningful test is whether projects such as pgrust can turn early development and performance claims into results that others can reproduce. That requires detailed benchmarks, continued testing, transparent handling of discovered defects and evidence about reliability across real workloads. The supplied material does not announce a release date, adoption milestone or formal evaluation for pgrust.
For developers considering similar work, Luu’s report leaves the practical next step clear: use the lower cost of experimentation to attempt more, then judge each result on evidence rather than the speed or scale of its creation. Whether these projects become useful tools—or contribute improvements to established software—will depend on the engineering and validation that follow the prototype.
Key Questions
What is the report arguing?
Dan Luu argues that faster AI-assisted development makes it more practical to try software projects that once seemed too large or time-consuming. He says that possibility should encourage more ambitious work, not replace testing.
What is pgrust?
pgrust is a Rust-based project related to PostgreSQL, developed by Michael Malis. Luu describes performance as its main pitch and says testing it has exposed bugs in PostgreSQL.
Does the report prove pgrust is faster than PostgreSQL?
No. The supplied material describes performance as pgrust’s main pitch but does not provide a comprehensive, independently verified benchmark comparison.
Are the projects described ready for general use?
Not on the evidence in the report. Luu says his personal experiments reached a “works for me” level, and that making them useful products or open-source projects would require more work.
What remains to be established?
Independent performance results, the status of the PostgreSQL bugs described, and whether the development-time gains can be repeated by other developers remain unclear from the supplied material.
Source: hn
