Agentically Observe, Understand, and Specify Software

In many large organizations, the documentation of production systems exists only in people's heads, in outdated functional concepts, or not at all. App2Spec is ReqPOOL's agentic consulting platform that observes existing software, understands its actual behavior, and derives a complete, evidenced as-is specification from it. This article shows how that works, what "evidence-first" means in practice, and in which situations it pays off.

The problem: systems nobody fully knows anymore

Banks, insurers, utilities, and public administrations run applications that have grown over decades. The original specification describes what was once supposed to be built, not what runs in production today. In between lie hundreds of change requests, workarounds, and special cases known only to the people who have looked after the system for years. When they retire or change employers, the knowledge leaves with them.

Yet exactly this knowledge is needed as soon as a system is to be replaced, put out to tender, migrated, or included in an application landscape analysis. The classic route is weeks of workshops, interviews, and manual documentation, with a familiar result: incomplete, subjective, and already outdated on the day it is finished.

For us, this is a question of what counts as an asset. Whoever owns the specification can have code regenerated at any time; whoever does not remains tied to grown code and the people who know it. For existing systems, "Own the Spec, not the Code" therefore means first of all: reconstruct the as-is specification systematically.

How App2Spec works

App2Spec works in four steps that build on each other:

1. Observe. An AI agent operates the target software on its own, like a user, or the platform records a real user at work. Every screen and every event is captured, including screenshots. Web applications and Windows desktop applications are supported; recording in the browser works without any local installation.
2. Analyze. The AI transforms screenshots and events into standardized entities of a knowledge graph: screens, controls, navigation paths, data fields, business rules, and process steps.
3. Specify. From the knowledge graph, App2Spec compiles a complete, consistent as-is specification, including derived business rules and end-to-end process flows with process diagrams.
4. Publish and export. The results are available on an online platform and can be exported to Word, PDF, and Markdown, in German or English, configurable per project.

The difference from classic reverse engineering of source code: App2Spec treats the system as a black box and documents what users actually see and do. That is particularly valuable where source code is not accessible, for example with heavily customized standard software or with systems run by external providers. For the inside view of source code we use SLIP, a platform of our own; the two complement each other.

Evidence-first: every statement is backed

A specification is worth only as much as the trust it earns. That is why every statement in an App2Spec specification is traceable to the underlying evidence: the screenshots and event logs from which it was derived. Anyone who doubts a business rule can see with one click on which screen and with which input it was observed.

This has three practical consequences:

• Objectivity. The software is documented as it actually is, not as it was once specified or as it is remembered. Knowledge gaps, outdated documents, and subjective assessments disappear.
• Auditability. For tenders, audits, and system replacements, evidenced statements are a decisive advantage: vendors, internal audit, and business departments work from the same traceable fact base.
• Comparability. Because the knowledge graph rests on a standardized ontology, specifications are structured uniformly across different applications. Landscape analyses and replacement projects benefit directly.

Observe software. Understand it. Specify it.

Typical use cases

In all four cases, the as-is specification is not the end but the beginning: it is the well-founded basis for target concepts, tenders, migrations, or agentic redevelopment, in other words for the core topics of ReqPOOL consulting.

No license costs, EU hosting, no training on your data

App2Spec is not software you buy. The platform is part of the ReqPOOL consulting service and is used by our consultants in projects. Three commitments follow from this:

• No license costs. In ReqPOOL consulting projects, clients pay no license fees for the AI tools we use. You pay for expertise and effort; the AI dividend stays with you.
• Data protection and compliance. The platform is hosted in the EU. Client data is not used to train AI models and does not leave the EU, not even for AI processing.
• Speed. The as-is analysis that classically requires weeks of workshops and manual documentation shrinks, according to our experience from ReqPOOL projects 2023 to 2026, from months to hours or days; the results can be validated per project.

The next step

If you want to replace a system, put it out to tender, or simply understand it again, an evidenced as-is specification is the first step. The App2Spec page lists features, compliance details, and frequently asked questions; how we analyze and document existing applications overall is described in our Reverse Engineering service. In an expert conversation we are happy to clarify which of your applications is suited for a start and what scope a first analysis would have.