Keboola uses an AI agent setup to remove legacy Zend code from its platform

·2 views

Keboola Connection still runs part of its system on Zend Framework 1 while the rest has moved to Symfony, and the team wants to remove the old code quickly. An AI agent was given a short brief with a goal and a Friday deadline, and in about 22 hours it merged 13 pull requests with three more open. The setup relies on a prepared dependency map, repository skills, and gating rules that decide when a bot approval suffices and when a human must review. Each change moves through a fixed path from draft and end-to-end tests to merge, with people stepping in only at review gates. The team describes the arrangement as a near-autonomous 'dark factory' for code migration.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why Software Engineering Still Matters as AI-Driven Vibe Coding Spreads

AI-assisted "vibe coding" now lets people build working web apps from plain-language prompts, making development far more accessible to founders and creators. Authors argue that as these apps grow to manage complex state, heavy traffic, and sensitive user data, core engineering practices such as modular design, automated testing, version control, and security modeling remain necessary. Skipping these standards can create serious problems, including hidden security flaws like OWASP Top 10 issues, risky third-party dependencies, codebases that are hard to maintain, and architectures that fail under concurrent load. The piece concludes that AI code generation is a strong productivity tool, but developers must still act as architects and reviewers to keep production systems stable and secure.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

SIP Trunking vs Cloud PBX: When an Organization Still Needs Its Own SBC

Once a company grows past a few hundred phones, its phone system choice becomes an infrastructure decision rather than an app selection. It must decide whether to run its own PBX and session border controller while buying SIP trunks, or move everything to a hosted provider. The article explains that the SBC is often overlooked in cost comparisons, since it protects the PBX from SIP scanning, translates between carrier and vendor dialects, manages call admission and failover, and anchors media. Keeping on-premises gear makes sense when the PBX was recently purchased, when it is tied to physical systems like paging or alarms, when contact center tools depend on it, or when a team with real expertise is already in place. The author, from a vendor that sells hosted PBX, acknowledges this bias while noting that retaining your own equipment is sometimes the better answer.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Ten small Windows tools that automate the tedious asset work in game jams

A small game developer explains that most game jam losses come from tedious chores like renaming hundreds of image files, not from weak ideas. The article shares a practical tip for each chore, such as packing sprites per scene, checking tile spacing, and using nearest-neighbor scaling for in-game art. Each tip is paired with a small Windows program the author built, covering sprite atlases, tileset slicing, animation previews, retro sound effects, bitmap fonts, palette recoloring, pixel scaling, batch asset processing, and dialogue validation. The tools are presented as a way to shorten the non-creative parts of jam work so developers can focus on the game itself.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

PostgreSQL 18 Changes That Matter for Django Apps

PostgreSQL 18 was released on 25 September 2025, and Django 6.0 followed that December with support for PostgreSQL 14 and higher. A key change is that initdb now enables data checksums by default, so pg_upgrade can refuse to proceed when the old cluster lacks them. Upgraders must either initialise the new cluster without checksums or enable them on the old cluster first, which is slower on large databases. The release also adds an asynchronous I/O subsystem that can speed up sequential scans and vacuum, though its defaults are small and it is best verified on the actual instance. Finally, the planner can now use multicolumn B-tree indexes in some cases where the leading column lacks an equality condition.

0 comments Read more at DEV Community