Technical Capability Document

geneworks.ai is an agentic platform that runs the full ServiceNow delivery lifecycle, requirement to release, rather than assisting a person through it. That distinction is the whole proposition. A code assistant makes an individual developer faster. geneworks.ai executes the delivery cycle itself: it plans the work, produces a reviewable technical design, builds the configuration in dependency order, tests it autonomously against both ATF and the live user interface, captures the evidence as it runs, and hands back a scoped update set for the customer’s normal promotion path.

The platform runs 152 specialist agents inside an SDLC-governed loop with two mandatory human gates: design review before build, and evidence review before deploy. Neither gate is configurable. Nothing reaches a customer instance without a named human accountable for the decision and the evidence to support it.

The platform carries more than 500 encoded ServiceNow skills — delivery patterns, standards, controls and testing rules captured once and reusable across every engagement. This is the mechanism by which an integrator’s methodology stops being documentation and becomes executable.

geneworks.ai is built outside ServiceNow deliberately. That choice lets it scale past the platform’s own execution limits, adapt to the customer’s standards rather than impose ours, and stay model-agnostic across Claude, OpenAI, Gemini and successor frontier models. The customer keeps full control of how and where AI touches their instance. Nothing is installed into the customer’s ServiceNow environment. The platform is delivered as a managed subscription running on a dedicated, single-tenant environment for each customer; Section 3 sets out the deployment options.

Alongside those figures, modelled pending a named reference engagement: 50–60% faster build cycles, 25–40% lower managed-services run cost and approximately 35% fewer human hours per engagement. A CMDB assessment that conventionally takes four to five weeks is modelled to complete in under an hour, and upgrade cycles from months to weeks, at roughly one tenth the token cost of comparable agentic approaches.

Every ServiceNow change runs the same cycle: plan it, build it, test it, sign it off, and find out later what broke. At enterprise scale, that cycle carries structural cost in five places. Each of the five is repetitive, rule-governed and verifiable, which is precisely what makes it automatable under governance.

For a systems integrator, the compounding problem is that all five costs scale linearly with client count. Every new account adds another instance to assess, another regression suite to run manually, another set of tribal knowledge to rebuild. Conventional tooling improves the unit cost of each task. It does not change the shape of the curve.

Six characteristics separate geneworks.ai from AI code editors, from AI software factories, and from conventional platform tooling.

Each zone has a distinct trust boundary. Customer content stays inside the dedicated environment.

Application services and customer data run within the customer’s own isolated node, encrypted at rest under customer-managed KMS keys.

Encryption. AES-256 at rest with keys held in a customer-managed KMS; TLS 1.2+ in transit. Because the keys are customer-managed, the customer retains the ability to render their data inaccessible independently of geneworks.ai.

The core operating pattern runs from a plain-language requirement through to a verified, deployed change. The two human gates are not configurable.

The platform’s reusable primitive is a Gene: the smallest reusable unit of executable enterprise intelligence. A Gene can be a policy, a delivery pattern, a control, a testing rule, a best practice or a decision standard. Genes move through a four-step cycle of encode, orchestrate, verify and evolve. In practice a standard encoded once for one engagement becomes executable across every subsequent engagement without re-teaching, and that is why the platform ships with more than 500 ServiceNow skills already encoded.

geneworks.ai is operated as a managed service. Nothing is installed into the customer’s ServiceNow instance, and the customer does not run the platform themselves. Each customer receives their own environment, provisioned and operated by geneworks.ai, reached over HTTPS with SSO through a multi-region delivery edge. Practitioners work in a browser; there is no desktop client and no infrastructure to stand up on the customer side.

The distinction that matters in an architecture review is that this is not conventional multi-tenant SaaS. There is no shared application runtime, and no shared database with row-level separation between customers. Each customer runs on a dedicated node with its own data stores, its own credentials and its own encryption keys. The only component shared across customers is the control plane, and it holds account, entitlement, billing and usage metadata alone.

geneworks.ai is offered in one deployment model: a managed, single-tenant environment on AWS, provisioned and operated by geneworks.ai. There is one such environment per client instance environment, and the customer needs no infrastructure of their own.

On-premise. On-premise deployment is not part of the standard offering. Where a customer has a regulatory or deployment requirement that the managed model does not meet, a dedicated arrangement can be discussed case by case.

Customer content is held inside the dedicated environment and does not leave it. The AI inference path is private and region-pinned rather than traversing the public internet, so prompt and response traffic stays within the nominated region alongside the data it draws on. Egress IP ranges are published for allow-listing where a customer restricts outbound access from ServiceNow.

Available hosting regions and the residency options attached to them: [TO CONFIRM]

Each customer receives a dedicated environment provisioned as part of onboarding. Provisioning covers the isolated node, the customer-managed KMS key configuration, the outbound tunnel from the delivery edge and the private inference path. This is the work covered by the one-time setup fee described in Section 8. Most engagements are connected and scanning within a single working session once the environment is in place.

Because the platform runs outside ServiceNow, product updates are applied by geneworks.ai to the dedicated environment. They require no ServiceNow upgrade, no update set and no change window on the customer’s side, which is the practical reason the platform can move at a different cadence from the instance it works on. Release cadence and customer notification lead time are set out in the support model at 7.2.

Because each environment is separate, updates are applied per environment rather than to a shared estate, so a release can be staged across customers instead of landing everywhere at once.

Isolation is achieved at the infrastructure layer rather than the application layer. Each customer receives a dedicated AWS node running the full application service set with its own data stores, its own credentials and its own encryption keys. There is no multi-tenant database with row-level separation, and therefore no class of application-layer defect that could expose one customer’s content to another. Cross-tenant learning does not occur: one customer’s encoded intelligence is never applied to another customer’s environment.

geneworks.ai authenticates to ServiceNow using customer-supplied credentials via OAuth 2.0, operating as a dedicated integration user under a least-privilege, scoped role set. The admin role is not required for normal operation.

The platform reads instance metadata, configuration records and, where required for testing, test data. Production business records are not required for the platform to operate and are not replicated wholesale.

This is typically the first question an enterprise security team asks. The answer is unambiguous, and it is enforced contractually rather than by policy statement alone.

Every agent action, human approval and deployment is logged with actor, timestamp and payload reference. Evidence packages are retained immutably against the run that produced them. Both human gates record the identity of the approving reviewer, which is what makes the delivery chain defensible under audit rather than merely well documented.

geneworks.ai supports development, configuration and automation across the core ServiceNow platform.

Commercially the platform is offered as two motions: Enterprise platform, and Professional services. Capability itself is packaged by depth. Each depth includes everything below it, so moving up adds capability rather than replacing it. Subscription is per named user. Term, seat minimums and rates are agreed jointly.

The mapping from each product — and from Ask Gene as a surface — to the specific cost it removes. Outcome ranges below are modelled.

Verify is the evidence half of SDLC automation. It executes functional, regression, UAT and process testing autonomously, and produces an audit-ready evidence package for every run. Coverage spans both ServiceNow ATF, for platform-layer assertions, and browser-driven functional testing that exercises the actual user interface and user journey. Self-healing tests absorb routine UI change without manual maintenance, which is what keeps a large regression estate viable across many client instances.

Command Center

A mission-control view of an entire instance in three layers: applications, connected ServiceNow components such as notifications and reports, and the metadata layer covering sys metadata, micro components, sys logs and sys audits. The map is spatial, zoomable and filterable per application. Functional tabs include Collisions (ask a natural-language question, receive blast radius and impacted applications), Activity (live tracker of in-progress work), Dead Weight (findings such as business rules never invoked by any active process), Tech Debt Calculator (the cost of maintaining customizations against the savings from returning to baseline, framed for an executive audience) and Who Can Do What (role and access analysis across tables).

Application Explorer

Each ServiceNow application is modeled as its own workspace with dedicated agents. Tabs are Overview (data model, analytics, business logic, UI, security and access, process automation), Ask Gene, Plan (high-level requirement through to full technical design document), Configure (the delivery loop), Testing (ATF and functional), Troubleshoot (root-cause agents) and Artifacts (functional requirements plus a versioned, auto-updating technical design document).

Artifacts

Generates delivery documentation directly from the work performed: functional requirement documents, technical design documents, test evidence reports and assessment readouts, across DOCX, PDF, PPTX, XLSX and HTML from a maintained template library. Technical design documents are versioned and update automatically as the underlying work changes, which removes the familiar problem of design documentation drifting out of sync with what was actually built.

The full PDLC loop runs from approved design to scoped update set, building in dependency order via the governed MCP connection. A built-in IDE supports direct inspection and intervention where a practitioner wants it. Because the output is an ordinary ServiceNow update set, the customer’s promotion and change management process applies to it unchanged.

Where automation is authored as code, geneworks.ai uses the ServiceNow Fluent SDK to express Flow Designer automation as TypeScript, producing artifacts that are version-controllable, reviewable in a standard code review process and diffable across releases. For a practice managing many client environments, that diffability is what makes it possible to demonstrate exactly what changed between them.

CMDB Rapid Assessment

The customer connects an instance and runs a scan; there is no consultant-led discovery workshop. Output includes an overall assessment score with health trending; scoring across completeness, correctness and compliance; analysis of mandatory against recommended and discoverable against non-discoverable attributes; CI inventory by class, data source analysis, IRE review and top findings; a health dashboard with table selection and prompt-generated compliance conditions; CSDM alignment maturity scored on a CMMI 1–5 scale with a phased remediation roadmap; prioritized remediation handed directly to Gene as workable tickets; plus scan history and PDF export. An Executive Readout presents a single score and maturity band where every figure drills through to its underlying measurement.

Catalog Center

Every catalog item is scored 0–100 across four equally weighted pillars — Build Quality, Experience, Governance and Workflow — with the composite shown as a scoring badge. Bands are 85+ Excellent, 70–84 Good, 50–69 Needs Attention, and below 50 Critical. Two views are provided: a builder-facing fix-it queue and a leadership-facing portfolio rollup. Item pages expose the score, the underlying assessment, an edit surface and recommended fixes actionable through Gene.

Upgrade Assist

Select a target release; the platform detects current version and patch level and scans the instance against the release documentation, a document set routinely exceeding two thousand pages. Findings are grouped as blockers, affected items, could-not-verify, manual review and informational. Coverage spans 900+ plugins, 200+ applications and 300+ tables. A consolidated Findings and Fixes view offers a Fix with Gene action that routes each item into the delivery loop and deploys, after which a re-scan verifies closure. Analysis history logs completed and failed runs.

Intelligent Normalizer

External normalization for CMDB and SAM, communicating with ServiceNow exclusively via REST API across four table domains: companies and vendors, software installs, locations and departments. It is backed by geneworks.ai’ own publisher and product database, refreshed nightly by a dedicated agent set, so no third-party normalization data subscription and no SAM Pro dependency is required. High-confidence matches auto-accept; low-confidence matches route to a review queue with accept, reject and escalate actions. Custom rules are supported. The platform reports total records, unique variants, canonical records and match rate, supports blanket-run deployment back to ServiceNow, exposes the publisher and product catalog including lifecycle data, and maintains run and audit history with schedulable jobs.

AI Optimization Advisor

A technical debt assessment surface producing a debt index, annual maintenance cost exposure, upgrade readiness percentage and modeled savings, supported by a debt breakdown, AI findings with severity and confidence ratings, business impact statistics, an ROI waterfall and a remediation roadmap timeline. It is built for the executive conversation: it quantifies the cost of the current state rather than describing it qualitatively.

The conversational layer across every product. It is a surface, not a fourth product. Ask the instance anything in plain English and Gene routes the question to the right specialist agents, answering, assigning tasks, brainstorming or running impact analysis, and acting on the answer where action is warranted. Because it draws on workspace memory and operates through governed workflows, the answers reflect the actual state of the environment and the actions remain inside the same control regime as everything else on the platform.

All communication with ServiceNow occurs over authenticated REST APIs through the ServiceNow Integration Service using the Model Context Protocol. This is the only path by which the platform reads from or writes to a customer instance. Changes are delivered as standard scoped update sets, so they follow the customer’s existing promotion path and change management process without exception. Nothing is installed into the customer instance, and nothing is deployed outside the customer’s established governance.

The onboarding checklist is deliberately short. Most engagements are connected and scanning within a single working session.

The platform is designed for a practice operating across many client environments simultaneously. Each connected instance is an isolated workspace with its own credentials, memory and evidence store. Encoded standards can be scoped to a single client or promoted to a practice-wide library where the standard is the practice’s own rather than the client’s. That promotion path is how a delivery method moves out of the handbook and into every engagement the practice runs.

Because each customer runs on a dedicated node, capacity scales horizontally by environment rather than contending for shared resources. One client’s heavy regression run does not affect another’s delivery throughput.

Concurrent instance limits, throughput per scan type and per-environment rate limits: [TO CONFIRM: current tested figures]

For a systems integrator relationship, enablement is delivered as practitioner training on the delivery loop and the gate model, Gene encoding workshops that capture the organization’s own delivery standards into the platform, and access to the workshop material set covering CMDB, HAM and SAM. The intent is that the customer’s own practitioners operate the platform independently rather than depending on geneworks.ai for delivery execution.

The economic point of the architecture is that delivery capacity stops being a function of headcount. A practice adds throughput by connecting environments and encoding standards, not by recruiting. The 500+ skills that ship with the platform are the starting position; every engagement adds to the encoded library, and that library is reusable across every subsequent account.

Consumption-based AI pricing makes delivery cost unpredictable, which in turn makes practitioners ration their use of the tool — exactly the opposite of what an organization wants once the platform is embedded in the delivery method. A flat per-user subscription, with unlimited AI usage under Fair Use on Professional services, removes that friction entirely. The cost of an engagement becomes a function of team size and duration, which is something a practice already knows how to price and forecast.

The one-time setup fee reflects the single-tenant architecture. Each client environment is a dedicated, isolated node with its own keys and its own connection path: real provisioning work, performed once per client instance environment, rather than a tenant record in a shared system.

Rather than present a fixed rate card, we would prefer to settle the licensing and commercial approach together. The right structure depends on the engagement model — internal use, reseller, or a broader strategic partnership — and on how many client instance environments are in scope. Section 5.2 defines what is included at each depth of capability; the rate card, seat minimums and term are the part we would work through jointly. There is no public rate card.

The product architecture is three products: Software factory, SDLC automation and Organizational Memory, with Ask Gene as the conversational layer across all of them. Plan, Build, Verify and Optimize remain the workspace headings underneath. Govern is the family under active expansion, extending the standards, controls, risk checks and approval logic that already run inside the delivery loop into a first-class surface of its own. Govern is roadmap, not shipped.

geneworks.ai was born in ServiceNow because that is where the delivery economics are most acute and most measurable. The architecture — encoded intelligence, agent orchestration, mandatory human gates, immutable evidence capture, single-tenant isolation — is deliberately stack-agnostic. The stated direction is to become the operating system for autonomous enterprise delivery across any enterprise platform, with ServiceNow as the proving ground rather than the boundary. The named targets beyond ServiceNow are Salesforce, AWS, Java, Angular and .NET applications.

The strategic read is that a relationship established on ServiceNow delivery extends naturally into the rest of the estate as the platform expands, rather than requiring a fresh evaluation for each stack.

A provisional patent application titled “Artificial Intelligence System for Continuous Enterprise Software Development and Governance” has been filed, covering the continuous delivery and governance architecture described in Section 2. Further detail is available under separate discussion.