Startups & Funding

Islo Raises $15 Million for Controlled AI Software Factories

Islo raised $15M led by Insight Partners to build controlled AI software factories — persistent microVMs, credential-scoped gateways, and a Databricks inference partnership.

Islo Secures $15 Million to Build Controlled AI Software Factories
Islo Secures $15 Million to Build Controlled AI Software Factoriesblickpixel / pixabay
By James Calloway5 min read

Why it matters

  • Islo secured a $15 million investment led by Insight Partners to expand its Enterprise Software Factory platform.
  • The platform runs agents in persistent microVMs with snapshots that preserve dependencies and database state.
  • A gateway supplies scoped credentials so agents never hold reusable API keys; all allowed and denied requests are logged.
  • A Databricks partnership provides model inference via Model Serving and scoped enterprise data access; Islo joined Databricks' Built on Program.
  • Islo is part of Incredibuild, which acquired workflow-automation firm Kypso in 2025.

Islo has secured a $15 million investment led by Insight Partners to expand its Enterprise Software Factory platform, wagering that the real constraint in AI-assisted development is everything that happens after an agent starts writing code. The funding, announced alongside a Databricks partnership, will support product development and adoption of the platform as a dedicated product line within the Incredibuild family of products.

The market logic behind the round is straightforward. AI coding agents can now generate patches at scale, but enterprises still have to test, secure, review, and deliver that code before it reaches production. Islo's pitch is that generation is only one step in a chain of engineering tasks — running the application, checking the patch against its dependencies, and showing a reviewer what happened are separate problems that current tools largely leave unsolved.

"AI can dramatically increase the amount of code an engineering organization produces, but that doesn't automatically translate into more software being delivered safely," said Adam Gold, Islo's chief product officer, in the announcement.

What is a software factory, in practice?

Islo describes its approach as a software factory. Instead of a developer initiating every task in a chat window, teams define workflows that respond to events or run on a schedule. The company says early customers are applying this model to:

  • refactoring
  • code review
  • quality assurance
  • bug fixes
  • issue triage
  • smaller feature requests

Each workflow carries gates for testing, functionality, security, performance, and cost. Engineering teams decide where human approval is required, and sign-off is retained before AI-generated changes reach production. In plain words: the agent proposes, but a human still decides what ships.

The promise of autonomous, repeatable development work is real, but the announcement says little about measured outcomes — no customer counts, throughput numbers, or defect-rate data appear in the release.

Why do persistent microVM environments matter?

The platform's foundation is Islo Machines — isolated development environments that can run an application's services alongside an agent. These are microVMs, lightweight virtual machines with their own operating-system kernels, rather than containers that share the host kernel.

A Machine can include dependencies, configuration, a browser, databases such as Postgres, and services such as Redis. That gives an agent a working application to exercise rather than asking it to reason about source files alone.

Persistence addresses a known limitation of short-lived sandboxes. Snapshots preserve installed dependencies and database state, so subsequent work resumes from a prepared environment. Jobs continue after a developer disconnects — which matters for tasks that take hours or span multiple iterations.

Islo also offers deployment inside a customer's virtual private cloud. For enterprises, the location of the runtime becomes part of the deployment decision alongside the choice of model and agent tooling. The takeaway: agents get real infrastructure to work in, not a disposable shell.

How does the gateway stop agents from holding API keys?

An isolated computer still needs controlled access to repositories, issue trackers, and other systems. Islo's gateway handles that boundary by checking outbound requests and supplying credentials only when an approved request leaves the environment.

In Islo's described workflow, an agent uses a placeholder binding rather than receiving the underlying secret. Gateway rules can restrict destinations, URL paths, and request methods. An optional model-based judge can assess request intent alongside those static rules.

This design separates permission to perform a task from possession of a reusable API key. An agent reviewing a pull request, for example, receives access scoped to that workflow instead of the same broad credentials a developer holds.

Allowed and denied requests are logged, with export to security monitoring systems. Those records let teams reconstruct which external services an agent contacted and which operations its policies blocked. One caveat sits in the announcement's own wording: the effectiveness of that boundary still depends on the permissions and rules an enterprise configures. In plain words, the gateway is only as tight as the policy behind it.

What does the Databricks partnership add?

The Databricks relationship adds inference — the process of running a model to generate a response — and access to the enterprise data agents need for real work.

According to Islo's Databricks integration page, customers connect an existing workspace through brokered credentials. Agents can use models available through Databricks Model Serving and request scoped access to the customer's compute and data.

Data processing remains on the customer's Databricks clusters, while Islo supplies the isolated agent environment and the controlled connection. The integration records requests and jobs for review, supporting workflows such as data engineering, analysis, and batch processing.

The release describes inference as the starting point for a broader product integration, including Islo's participation in Databricks' Built on Program. That connects the software-factory strategy to an existing enterprise data platform instead of requiring teams to duplicate their data in a separate agent tool.

Where does Incredibuild fit in?

Incredibuild contributes distributed execution and caching technology to the platform. Distributed execution spreads suitable build or test work across compute resources; caching reuses previous outputs when the relevant inputs have not changed. Both shorten the feedback cycle between an agent proposing a change and a team learning whether it works.

The relationship follows Incredibuild's acquisition of Kypso in 2025. Kypso's engineering agents and workflow automation drove Incredibuild's expansion beyond build acceleration into a broader software delivery platform — the lineage from which Islo now emerges as a product line.

For Islo, the $15 million backs the infrastructure around autonomous engineering: environments that stay available, permissions that can be enforced, and checks that connect generated code to release decisions. The opportunity, as the company frames it, is making repeated development work dependable enough to delegate while keeping engineers responsible for what ultimately ships. Whether that holds at enterprise scale will depend on evidence the announcement does not yet provide.

Source: unite.ai

Share this article:

More from James Calloway

James Calloway

Show full bio

News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.

722 articles

Related articles

  1. Islo Raises $15M to Scale Its Enterprise Software Factory
  2. AMD Says AI Agents Now Auto-Fix 75% of Radeon Software Bugs
  3. AgentKit, Expanded Evals, and RFT for Agents Ship Together
  4. OpenAI to Acquire Promptfoo and Build Red-Teaming Into Frontier
  5. Vijil CEO Vin Sharma: AI Agents Need Trust Infrastructure Before Autonomy

« Previous articleNext article »