Prelint

Prelint: The Essential Product Review Layer for AI-Driven Development

In the rapidly evolving landscape of software development, AI agents are now capable of generating vast amounts of code in seconds. However, while AI is exceptionally fast, it often lacks a deep understanding of human intent, complex economic models, and regulatory constraints. This gap leads to what is known as product drift. Prelint is the professional solution designed to close this gap by providing an automated product review for every pull request, ensuring that AI-written code remains strictly aligned with your established product specs.

What’s Prelint?

Prelint is an advanced AI-powered tool that reviews every pull request against your specific product specs to catch misalignment before it ships to production. While traditional tools focus on whether code is technically sound, Prelint asks, "Should this code exist?" and "Does this match our business intent?"

By treating your product documentation as machine-readable constraints, Prelint acts as a guardian for your product's integrity. It ensures that as AI agents iterate on your codebase, they do not accidentally introduce financial risks, compliance violations, or strategic deviations. In essence, while tools like Greptile handle the technical code review, Prelint handles the product review, creating a comprehensive safety net for modern engineering teams.

Key Features of Prelint

1. Automated Product Review for Pull Requests

Every time a developer or an AI agent opens a pull request, Prelint automatically triggers a check. It compares the proposed code changes against the documentation stored in your repository to ensure consistency and completeness. This process requires no manual meetings, allowing development to continue at high speed without sacrificing quality.

2. Machine-Readable Product Constraints

Prelint understands the full context of your product. It can detect contradictions between different specifications, identify conflicting technology choices, and flag duplicated efforts. By turning your product specs into active constraints, Prelint ensures that your business logic is never compromised by automated code generation.

3. Stakeholder Query Support (MCP)

With Prelint, stakeholders and non-engineers can get grounded answers to questions like "Why did we build it this way?" without interrupting the engineering team. This feature uses your actual product specs to provide context-aware answers, keeping engineers in their flow state while keeping the rest of the organization informed.

4. Enterprise-Grade Security

Security is at the heart of Prelint. The platform operates on isolated, per-organization infrastructure.

• Zero Retention: Data is isolated per organization with no sharing between tenants.
• No Training on Your Data: Your proprietary code and specs are never used to train any models.
• Least Privilege: Prelint only requests the GitHub permissions it absolutely needs.
• Encrypted Everywhere: All data is protected by TLS 1.3 in transit and AES-256 at rest.

Six Ways AI Code Drifts from Your Product

Without a dedicated product review tool like Prelint, AI-written code can quickly diverge from your original vision in six critical ways:

1. Business Logic Drift: AI may change pricing, discounts, or billing rules without understanding the financial impact. For example, an agent might switch pricing from a vendor currency to a customer currency, silently creating massive FX exposure and revenue leaks.
2. Compliance Risks: AI might store sensitive data, skip consent flows, or break retention rules. An agent adding a user activity log might store IP addresses without a proper retention policy, unknowingly violating GDPR and CCPA.
3. Tooling & Infrastructure Fragmentation: AI often introduces new vendors or dependencies simply because it can. This leads to paying for redundant services, such as adding Twilio when your company has already standardized on AWS SNS.
4. Domain Language Confusion: AI often uses generic industry terms instead of your established internal language. This splits concepts that should be unified, leading to a codebase that uses "merchant," "seller," "vendor," and "partner" to describe the same entity.
5. Scope Creep: AI agents may build features that were never requested, such as implementing a full i18n framework for a US-only product or setting up complex infrastructure before it is needed.
6. Strategic Drift: AI might build off-roadmap features or public APIs for tools meant to be internal, wasting multiple sprints on work that does not align with the company's long-term goals.

How to Use Prelint

Integrating Prelint into your existing workflow is seamless and takes only minutes. Here is how it works:

Step 1: Store Specs in Your Repo

Your product specs, business constraints, and compliance rules live directly in your GitHub repository. Prelint can read these files in common formats like Markdown or YAML. Because they are version-controlled, they are always current.

Step 2: Agent or Developer Opens a PR

Whether it is a human contributor or an AI agent shipping a new feature, they open a pull request as usual. Prelint activates automatically the moment the PR is created.

Step 3: Receive Product Review in Seconds

Prelint reviews the PR against your full product context. Any drift, conflicts, or gaps are flagged inline. For example, if a PR attempts to store charges in customer currency when the spec requires vendor currency, Prelint will flag the conflict and suggest a fix immediately.

Use Case: Preventing Financial Risk and Managing Roadmap Priorities

Real-World Pricing Correction

Consider a scenario where an AI agent attempts to modify charge-calculator.ts. The AI might write code that converts fares at today's rate and stores them in the customer's currency. Prelint would identify that this creates FX risk for advance bookings. By referencing the "Pricing: International Settlements" spec in Notion, Prelint suggests the correct fix: "Always store the charge in vendor currency. Show the customer a converted estimate at booking time."

Unblocking Enterprise Deals

In another use case, a stakeholder might suggest adding Google login to unblock enterprise deals. Prelint can jump into the conversation, referencing existing security specs to explain that the specific clients (like Meridian Health) require SAML/OIDC federation (Okta), not social login. This prevents the engineering team from wasting two weeks building the wrong solution.

Frequently Asked Questions

What does Prelint actually review?

Prelint reviews the intent of the code rather than the implementation. It checks pull requests against product specs, compliance rules, and business constraints to ensure the code should actually exist.

How is this different from code review?

Traditional code review checks if the code compiles, follows patterns, and is secure. Prelint provides a product review, ensuring the code matches the product spec, respects business constraints, and aligns with the roadmap.

Is my data secure with Prelint?

Yes. Prelint uses isolated per-organization infrastructure, has a zero-retention policy, and never trains its models on your data. It operates on the principle of least privilege for GitHub permissions.

What can non-engineers do with Prelint?

Non-engineers can use the MCP feature to ask questions about the product's architecture and history. They receive answers grounded in the actual product specs, allowing them to understand the "why" behind technical decisions without needing a meeting with engineers.