I don’t get the future of AI(-assisted) coding for PHP CMS projects?

Web development with a PHP/SQL CMS used to be (rather) simple:

• One developer, one laptop
• A locally checked out project, eg. via DDEV (Docker)
• An IDE like PhpStorm or VSCode, typing or pasting / remixing code snippets from stackoverflow
• Developer working in his/her office / working hours
• If it was a good day: Some kind of flow-state achieved
• Pushing changes to a staging or production system
• Closing the laptop at the end of the shift

Now there are these AI challenges of …

1. Not getting distracted in „idle times“, waiting for AI coding sessions to be executed / needing the developers feedback
2. Handling multiple, parallel coding sessions in your head – maybe even across different projects
3. Verify and adjust changes made by looking at the final result, e.g. local preview like https://my-project.ddev.site
4. The (theoretical) possibility to start long running tasks, which extend above the working hours

Joshua Morony described the first two challenges quite well, his current solution was a better notification management system (locally) while working with tmux:

Some commentators pointed out, that herdr could be another.

The big question for me is:

If coding work does not end in office / working hours, we need to move (some of it) into the „cloud“?

Even before AI, there was the attempt to move the development workflows to remote environments / the „cloud“, e.g. by GitHub: GitHub’s Engineering Team has moved to Codespaces (2021). Mainly because the struggle of setting up their complex project on local laptops and coming up with „nuke all“ scripts.

As far as I can tell, PHP agencies and freelancers didn’t go on that path and kept their local development enviroment on their machines. E.g. via DDEV, Docker (Compose), Laravel Herd, XAMPP, MAMP etc.

I would assume that most PHP CMS or framework developers still use Claude CLI / Desktop, Codex, Open Code, etc. on their local machine – or am I mistaken here?

Friction I: Bridging the gap between remote and local?

Seems to be that the goal would be „Project preview seamlessly available in local and remote sessions, ready to be verified by the developer as well as the AI agent?“

With claude code desktop and the browser extension (running in a separate chrome profile), if often used this working on local projects:

Verify these changes on https://my-project.ddev.site

But how could I use this frictionless, if I move some AI sessions to an remote environment, e.g. herdr on a VPS (or simply claude code cli / codex cli installed on a VPS).

DDEV announced a selfhosted coder.com instance recently (experimental):

• https://ddev.com/blog/coder-ddev-com-announcement

But as far as I can tell, there is no „seamless bridge“ to connect a local with a remote project. You can of course check out the same feature branch on coder.ddev.com – and as well in your local environment and set both up, pulling/pushing changes. Which still leaves you with a local IDE and a remote IDE.

This business insider headline captures the friction quite well:

Friction II: Parallel AI agent sessions for one DDEV project

This is one of the frictions, rather simply solved since DDEV – and tools like Claude Code do have support for git worktree:

• https://ddev.com/blog/git-worktree-contributor-training/

Kanban-Style (for orchestration)?

I’m still curious about the approach introduced by VibeKanban, with conflict resolution in mind. But I think it’s hard to move developers to other meta-tooling when providers like Anthropic, Open AI push for their CLI / desktop apps; or I might be mistaken here as well.

Interesting video for further context in terms of provider harnesses vs. third party tools: A problem for open AI agents like Pi – Maximilian Schwarzmüller

I saw a similiar mention of Hermes Agent’s Kanban feature.

A big question in my head is

• if one developer orchestrates the work
• or multiple devs – or even PMs and other roles – writing specs in markdown, triggering the execution of agents for first drafts (cultural shift)

Another meta tooling seems to be warp.dev for orchestration.

Another recent project mentioned is Orca – Agent Development Environment (ADE) https://www.onorca.dev/ (open source). It relies on (isolated) git worktrees – and can also run remote sessions as well as be selfhosted remotely.

„Orca is designed for people who already write code for a living and want to use AI as leverage — not as a replacement. It assumes you read diffs, care about commits, and keep a worktree tidy. If you’re looking for a no-code tool, Orca is not that.“

„You want agents to run remotely — over SSH, on a self-hosted Orca server, or in an on-demand VM — without giving up your IDE.“

Check their docs, multiple ways to run it: https://www.onorca.dev/docs/ways-to-run

… and there are a dozen other (open or commercial) tools?

How do agencies and freelancers adapt?

But I do wonder if small to medium sized agencies or freelancers will just stick to their well known workflow, slightly adapted:

1. one developer as orchestrator of his/her AI coding sessions, tasks are assigned as before by PMs, etc. (= accountability)
2. finding a way to integrate long running tasks after working hours / over night (= the tough cultural shift?)

I just can’t imagine that agencies won’t (have to) come up with a solution for long running tasks over night, because of the tough (cost-)competitive field of work they are in.

Samuel Reichör’s field work

One developer trying to tackle this challenge for agency work – with a dashboard and workflow approach – is Samuel Reichör from Linz / Austria: His project is called knecht.works (early development).

I haven’t seen many similar projects – my guess is because most developers are currently happy with their productivity boost achieved with Claude / OpenAI / Open Code tooling on their machines for now – compared to pre-AI-agent times?

What is your take?

How do you think about the future of web development – working with Craft CMS, Drupal, TYPO3, Laravel, etc. and AI tools?

Let me know on Bsky, Mastodon or at DDEV Discord. Cheers!

PS: Needless to mention that there is another challenge in terms of energy consumption and sustainability of using AI tools for coding. Make sure to follow the various Green Web movements!

Beitragsbild: Image by tookapic from Pixabay

Reactions

• A developer briefly suggested „portless + tailscale“ – I need to check if this would be compatible with DDEV projects and how these workflows would look like. (Thanks, Tobias!)
  • portless shouldn’t be needed, since DDEV does the reverse proxy stuff already
  • There is a plugin for tailscale: https://ddev.com/blog/tailscale-router-ddev-addon/
  • I currently assume that would enable local agents to access remote projects URLs (on a VPS) – or vice versa, remote running agents to access a local project URL.
• Orca – Agent Development Environment (ADE) https://www.onorca.dev/ added (Thanks, Thomas!)