The AI delivery gap, and what Domino’s latest release does about it

A data scientist opens a coding assistant and has a working application by five o'clock. An analyst points the same assistant at a dataset and builds an agent that drafts a first-pass summary for review. Both are common now, including those companies with a model risk function and an audit committee. What happens next is where things get difficult. The application sits in review status for months while someone reconstructs who approved the data it touches, and the agent keeps running on a laptop because nobody has decided what approving an agent even means. Teams can build in hours, but getting AI into the hands of business users safely remains the true challenge.

This disconnect between technical speed and business delivery is industry-wide. Domino's 2026 Enterprise AI Report (BARC/Domino, 639 senior AI leaders at enterprises above $100M) found that 93% of AI leaders report improved ability to move models from experimentation into technical production. Yet 57% report that ROI is growing no faster than their investment, and 44% expect regulatory sanctions from ungoverned outputs. Value remains trapped in governing that final delivery to the business, long after technical deployment is complete.

Domino 6.3 addresses that gap across the entire development lifecycle, spanning how work gets built, how applications reach business users, and how platform assets are governed over time. Coding assistants, extensions, and HPC workloads now land in a single system of record, giving compliance teams continuous visibility while data science teams keep building at full speed with an improved app building process.

Integrating coding assistants into the platform

GitHub Copilot, Claude Code, and Codex now ship preinstalled in the Domino Standard Environment. That reads like a convenience feature and behaves like a governance one. A coding assistant on a data science workstation usually arrives as shadow IT, running as a personal install invisible to the platform and to anyone later auditing what touched which dataset. Running inside Domino brings those tools into the governed workspace. This gives developers the speed of generative coding tools without creating compliance blind spots or manual audit headaches after the fact.

Teams needing a clean environment can start from an opt-out image. Anyone preferring a different assistant can bring their own tool into the same environment and route requests through the LLM Gateway, keeping every prompt audited. An assistant's workspace, settings, and chat history stay intact across workspace restarts.

Domino Skills give assistants platform-aware actions across datasets, jobs, and registered models. When asked to retrain a model, an assistant logs the run through Domino's experiment tracking instead of writing to an unmonitored notebook. The assistant’s work lands in the same audit trail as human development.

Delivering apps and platform extensions

Getting a model into production is a different problem from getting it into the hands of the person who has to act on it. The Enterprise AI Report highlights this gap, showing that 40% of organizations depend on a mediated path, such as a scheduled report or a request to a data scientist, to get AI output to a business user. Domino addresses this gap first by expanding platform capability directly inside daily practitioner workflows through Extensions. Developers can embed custom web applications and interactive tools directly into dataset pages, project sidebars, file context menus, model pages, and the admin panel. Because Extensions run on App Hub, every custom UI tool inherits standard Domino access controls automatically, eliminating the need to manage a second permission system.

That same platform foundation also streamlines how applications reach business stakeholders. App Preview lets builders and reviewers test a working draft on an author-only URL rather than reviewing static descriptions or slide decks. The evaluated asset is the working application itself. Publishing and deployment are now separate steps as well. Publishing pins the exact commit and records configuration immutably without starting the app, which shortens review cycles and ensures approvals attach to a specific release rather than a moving build target.

Gated Deployment allows admins to require approvals before an app, model endpoint, or agent goes live, writing every rule evaluation to the audit log as its own artifact. Gated deployment blocks unauthorized releases automatically at the exact moment business risk exists. This leaves data science teams free to build uninterrupted during development while giving admins control over production releases.

Unifying HPC and platform assets in one system of record

The Governance Center now governs assets continuously. Policies can enforce periodic revalidation, which is a recurring review anchored to a fixed calendar date. Expired bundles surface as Pending Expiration before lapsing. Policy versioning allows admins to publish mandatory or optional updates without revoking prior approvals. This prevents minor policy revisions from disrupting existing, compliant production models or triggering months of re-approval sweeps.

Beyond software policies, Domino extends this same system of record to heavy compute infrastructure. HPC on Slurm integrates high-performance compute workloads that previously ran outside the system of record. Domino hosts and governs the Slurm cluster inside your environment, letting existing scripts submit unchanged while compute scales dynamically with demand. This cuts idle compute costs while bringing heavy workloads into the unified audit trail.

Governing all of these assets in one place also solves a catalog discovery problem. Knowledge Manager structures catalog discovery through Tags and Properties that feed directly into platform search. Tags cover fixed attribute sets, whereas Properties capture variable asset details. Admin or Librarian roles define the namespaces and properties, while individual asset owners populate the specific values. This enables teams to find and reuse existing apps, datasets, or models instead of rebuilding them from scratch.

Getting started with Domino 6.3

Domino 6.3 is generally available on Domino Cloud and for self-hosted deployments. The 6.3 release notes carry the full change list, and if you self-host Domino, your account team can walk the upgrade plan through with you.

Upgrading to Domino 6.3 unlocks all of these capabilities across your platform immediately. To get the fastest return on your upgrade, admins should focus their initial rollout on three operational priorities. First, transition practitioner teams to the new standard environment so coding assistant usage is automatically captured in your audit trail. Second, apply periodic revalidation to policies governing legacy models so older deployments stay compliant. Third, attach deployment gates to agent and app workflows before teams begin requesting production compute.

If you need help designing governance across complex workflows that span multiple teams, Domino Solutions can co-build that architecture alongside your own engineers.