Skip to content

Product overview

AI IT House 26.10.5 is a provider-neutral, locally installed IS/IT knowledge and delivery framework for enterprise teams using AI-assisted tooling. The release supports Claude Code and Codex / OpenAI-oriented execution surfaces while keeping the core role model, workflows, artefact contracts, JAI semantics and governance rules provider-neutral.

The 26.10.4 Unified Lifecycle GA baseline completed the transition from several parallel domain models to a single normative Justifiable AI core with scoped Artifact, SABSA, EA, Publication and Runtime profiles.

What problem it solves

Enterprise AI-assisted work commonly fails in several distinct ways:

  1. the task starts with the wrong specialist viewpoint;
  2. work skips lifecycle or governance gates;
  3. generated content is persuasive but weakly evidenced;
  4. authorship, review and authority collapse into one model conversation;
  5. publication loses traceability to the semantic decision;
  6. later nobody can reconstruct which requirement, evidence or approval justified the use;
  7. a failed tool call or missing observation is silently converted into a confident-looking answer;
  8. a system retries or rephrases until something appears acceptable instead of escalating non-convergence.

AI IT House addresses these failure modes through:

  • curated role knowledge and explicit tool tiers;
  • 24 canonical enterprise workflows;
  • typed Artifact Contract v3 and a 28-class artefact taxonomy;
  • the six-layer JAI architecture with explicit human authority boundary;
  • append-only traceability and bounded supervision for consequential engagements;
  • SABSA semantic modelling and scoped EA/Open Exchange projection;
  • publication projection, render evidence and visual QA;
  • managed policy, reviewer isolation and provider parity;
  • A–H enterprise release qualification.

Release scope at a glance

Consult the release inventory for the generated package counts and the commercial implementation matrix for the extension's coverage and qualification limits. The release manifest is authoritative over historical or explanatory counts. Source availability, passing repository checks and qualified live deployment are separate states.

The normative architecture

The JAI core has six responsibility layers:

  1. Intent and Context;
  2. Meaning and Domain Model;
  3. Reasoning, Policy and Control;
  4. Solution Design;
  5. Realisation and Tooling;
  6. Execution and Evidence.

L0 Human Authority is external. Agents may prepare evidence, recommendations and dossiers, but they cannot manufacture L0 authority.

For consequential work, /layered operationalises the downward/upward cycle with typed contracts, independent review, bounded recovery and release gating. Begin with How to use JAI to choose the assurance arrangement and follow a complete example. Running a layered engagement explains its operating cycle.

Supported working domains

The catalogue covers architecture, business analysis, service design, software engineering, cloud and infrastructure, operations, governance/risk/compliance, cyber security, SABSA security architecture, data/AI, testing, delivery and sourcing.

What “Unified Lifecycle” means

The release does not mean that one AI model autonomously owns the entire lifecycle. It means that the major production subsystems share JAI semantic identity, release namespace and traceability rules:

  • Artifact Contract v3 writes current JAI-bound artefacts;
  • SABSA objects are profile-native at source;
  • EA is an architecture-domain projection, not a competing global truth store;
  • Publication requires a projection manifest and carries JAI lineage;
  • Runtime Agent/Policy/Execution records are JAI-bound;
  • historical 26.10.0–26.10.3 data remains forward-migratable.

What it is not

AI IT House does not replace accountable human roles and does not guarantee arbitrary natural-language truth. A structurally complete, well-reviewed record can still rest on a bad requirement, incorrect source observation or mistaken expert judgement.

The platform also does not claim that every possible direct role invocation is automatically forced through a /layered episode. The full Fabric supervision mechanism is strongly enforced once that path is used. Organisations requiring layered-by-construction for every consequential task should enforce that classification/routing at their entry boundary.

Why “Universal” is fair but bounded

The platform spans a broad set of IS/IT working contexts and output types. “Universal” should be read as broad enterprise applicability and shared governance/runtime mechanisms — not as unlimited autonomy, infallibility or automatic suitability for every environment.