Submit an idea
OpenEdge ABL Developer Pack · The vision

From the old Developer Studio
to a fully automated AI pipeline.

We bring every capability of the classic OpenEdge tooling into VS Code — for the people who work by hand, the people who work side-by-side with AI, and the teams ready to let AI agents orchestrate an entire feature end-to-end, on Windows and Linux.

See the pipeline → Submit an idea
Built for everyone

Three ways to work — your choice

You don't have to use AI to benefit. The same operational tooling powers all three tiers, so you move up only when you want to.

1 By hand

Full manual tooling

For developers who prefer hand work. Every database, PASOE, health-check, UI and API task lives in a real VS Code UI — no AI required.

2 AI-assisted

Semi-automated

AI sits beside you and asks the right questions to get something done — you stay in control and approve each step.

3 Fully automated

AI orchestration pipeline

Agents drive the whole feature — spec, generation, business logic, UI, validation and deploy — handing off to one another with the right specs.

Where we are today: putting the plugins through their paces and getting the hands-on AI tier right. The orchestration tier follows naturally — it leans on patterns we've already run on other projects.

The real destination: move away from the lines of code and toward described features, where humans and AI agents collaborate to drive any change to done.

Every agent reports to one supervisor, who routes the work — with the right specs — to the next agent in the pipeline. A coordinated team, not a single prompt.
Worked example

Extend a React app with Customer data from OpenEdge

It starts with an intake: a business & technical analysis the AI performs side-by-side with your codebase, which turns the requirement into a plan and the OpenAPI spec. From there each stage can run by hand, with an AI assistant, or fully via MCP. Across it: intake & analysis is AI + you, the middle runs on AI (where the ABL backend and React UI build in parallel), and you approve the deploy at the end. (If the analysis finds a schema change is needed, it adds the field too — otherwise that step is simply skipped.)

AI + you AI You approve Intake & analysis — business + technical, AI side-by-side with the codebase Turns the requirement into a plan: what to build, which APIs, and whether a new database field is needed OpenAPI spec · Customer By hand in the UI — with AI — or via MCP OpenAPI plugin running in parallel BACKEND · ABL FRONTEND · React SAME PATTERN SCALES — OPTIONAL EXTRA LANES Mobile app · Flutter Another frontend track on the same spec + widget / unit tests Extra service · batch / integration Another backend track, same pattern + ABLUnit Branch from the same spec · fold into the same E2E + approval Add the required field Conditional · only if the analysis asks Data Administration Generate server ABL server stub from the spec OpenAPI generator (MCP) ABL business logic AI + OpenEdge MCP knowledge AI assistant ABL unit tests Verify the logic in isolation ABLUnit (MCP) Performance check Is the new API fast enough? HCK (MCP) ABL object-leak check No leaking objects on PASOE PASOE (MCP) Schema + delta .df Ship the field change · incremental Data Administration (MCP) Build with PCT Compile ABL to r-code PCT (MCP) Generate client Typed API client from the spec OpenAPI generator (MCP) React UI AI + framework & design guidelines UI (MCP) React unit tests Verify components in isolation Vitest · RTL (MCP) frontend ready · waits for the E2E End-to-end test Frontend + backend together · the whole feature E2E suite (MCP) You review & approve → Deploy Publish to PASOE · the targets it needs Deploy MCP servers & packages Guideline-driven incremental delivery React app now reads Customer data — shipped, tested, verified
Analysis / intake Manual / UI AI-assisted MCP / automated Build & deploy Dashed box = backend ∥ frontend run in parallel — each track has its own unit tests, then one E2E test before you approve. Left rail = who's mainly involved (AI + you → AI → you approve) Faded panel = optional extra lanes the same pattern supports (e.g. a mobile frontend, another backend service) — not part of this run.
AI orchestration

A squad lead coordinates a team of agents

The AI side of the example above. A squad lead plans the work and hands each task to an agent. The agent does it and reports back to the lead — then the lead hands the next task to another agent. Agents never talk to each other directly; everything routes through the lead. That's the whole point of the squad: one coordinator, clean hand-offs, and it scales — add more agents and the pattern doesn't change.

Squad lead plans the work · hands off · collects results every hand-off goes through here Analyzer agent intake · plan · what to build Backend agent — ABL server + business logic spawns its own subtasks Frontend agent — React API4UI · React UI spawns its own subtasks Validator agent HCK perf · PASOE leak · tests + more agents — same pattern
Two-way hand-off: lead ↔ agent Agents never talk directly — every task and result routes through the lead. Scales by adding more spokes (agents), not more connections between them.
How it's possible

Logic never lives behind the UI

A different axis from the three tiers above: this is how each capability matures so an AI can reach it — moving through four phases and living in a headless core that the extension, a CLI, and an MCP server can all call.

The four phases

1 · Manual 2 · CLI / headless 3 · Skill 4 · MCP Anything a human can do, an AI can eventually do via MCP.

Layered architecture

AI / Agent VS Code ext MCP server CLI optional · per plugin Headless core logic lives here — no VS Code required via MCP
The toolkit

One pack — every operational capability

The AI4YOU operational tooling layer on top of the Progress OpenEdge language & MCP foundation. Each plugin is a standalone capability — usable on its own, and exposed to AI through MCP.

AI4YOU plugins

Data Administration

Manage OpenEdge schema — tables, fields, indexes — and produce incremental delta `.df` files.

OpenEdge DataDigger

Browse, query & edit database data and explore the schema.

OpenEdge ABL – OpenAPI

Generate ABL servers and typed clients from OpenAPI specs, and keep them aligned.

API4UI – UI Designer

Visual drag-and-drop UI designer for OpenEdge ABL & modern frontends.

HCK – Health Check Kit

Database health-check monitoring & performance analysis.

OpenEdge PASOE

Configure & manage PASOE servers — inspect agents and catch leaking ABL objects.

PCT build (wrapper)

Wrapper around PCT — ships with the Progress OpenEdge install — exposing all PCT capabilities through MCP for full automation (compile ABL to r-code and more).

Config Management

Central owner of shared config, scaffolding, compile/run & MCP lifecycle.

Progress

OpenEdge AI assistant

AI assistant for OpenEdge — used to write & refine ABL business logic.

Quality bar

Targeting the canon

Our goal is to measure everything the pipeline produces against the “canon”: the canonical set of standards and guidelines that defines good. We're still working towards covering them all — it's an ongoing effort, and the analysis is honest about what's enforced today versus still on the way.

Security

OWASP

Generated APIs and code are weighed against OWASP guidance (e.g. the Top 10) so common vulnerabilities are designed out from the start — not patched in after a breach.

Product quality

ISO/IEC 25010

The international model for software product quality — functional suitability, performance, security, reliability, maintainability, usability, compatibility, portability. Our check-iso-standard pass scores work against it.

Consistency

Our own guidelines

The house canon of ABL, API and UI conventions plus the current frontend framework & design rules — so generated output looks and behaves like it belongs in your codebase.

“Canon” = the agreed reference set we want every agent and generator held to. A work in progress — we're rolling these in step by step, transparently, aiming for predictable quality rather than box-ticking.
Shape the roadmap

Submit an idea

We're here for everyone. Tell us what's missing — whether it's a hand tool, an AI assist, or a new pipeline stage.

Goes straight to our team
Leave your email if you want — we'll personally reach out and let you know when we pick it up and intakes start. No address, no worries: it still lands with our team.
Thanks! Your idea has been sent.