Agent Control

Agents open 40% of pull requests. We check what they ship.

A fixed-price audit of agent-written code: architecture drift, blast radius, missing tests, and what your agents can reach. Every finding is traced to a commit.

Why now

In September 2026, more than 40% of pull requests on GitHub started with an agent's first draft, up from 17% in March.

By 2028, 70% of the world's 2,000 largest companies will continuously validate what their AI agents decide. Today 14% do.

IDC FutureScape 2027

By 2028, 75% of contracts with technology vendors will carry liability terms for what AI agents do on their own.

IDC FutureScape 2027

When teams call us

A failure after an agent's change

A regression or an incident, and it took days to find the cause.

The codebase is drifting

Duplicate logic, module boundaries bypassed, tests that do not cover what agents change.

A customer or an investor asks

A security questionnaire or a due diligence asks how AI-written code is controlled.

Agents can reach too much

CI tokens, secrets, MCP servers and agent account permissions that nobody has reviewed.

Three fixed-price steps

1 week01

Blast Radius Audit

$6,500

How far an agent's change, and an agent's credentials, can reach.

  • Architecture graph of your repository and where agent-started changes land in it
  • Boundary violations, blast radius against test coverage, duplicate logic, new dependencies, merges without human review
  • What agents can reach: instruction files, MCP servers, secrets, CI tokens, account permissions
  • Report with every finding tied to a file, a commit and a pull request; fix plan with effort; a one-hour readout
  • Read-only repository access. No production access needed.
2-3 weeks02

Guardrails Sprint

$13,000

We fix the top findings and put the limits in place.

  • Architecture rules as code, checked in CI
  • Blast-radius check on every agent pull request
  • Repository instructions and skills so agents follow your architecture
  • Human review threshold by blast radius; no merge without a test
  • Least privilege for agents: scoped short-lived tokens, an MCP allowlist, secrets out of reach
  • Evidence report tied to git: finding, code, commit, test, retest
2 days a month03

Drift Check

$2,600 / month

A monthly recheck so architecture and permissions do not drift.

  • Audit metrics rerun every month
  • Review of new agents, tools and MCP servers
  • Answers to customer questionnaires about AI-written code
  • Evidence kept current

Prices exclude taxes. No VAT is charged to clients outside the EU.

How it starts

  1. 1

    Run Wingbox on your repository. It is open source and runs locally.

  2. 2

    Book 30 minutes. We read the result together.

  3. 3

    Blast Radius Audit at a fixed price.

Why Aileron

  • The engine is public: Wingbox (MIT), a code graph and blast-radius analyzer for AI coding agents.
  • Evidence you can verify: every finding and every fix is tied to your git history.
  • 20+ years of architecture: systems handling over 1 million calls a week and 1.5 million orders a month.
  • Zero-trust, certificate-based PKI securing live in-flight aircraft data streaming to the cloud. The same engineering now scopes what your agents can reach.
  • 2nd place, solo, at RAISE YOUR HACK 2025, against 220+ teams.
  • Open-source contributor to Archestra, an enterprise AI platform with an MCP gateway and guardrails: built its Browser MCP feature (merged January 2026).
  • Organization approved in Anthropic's Cyber Verification Program (Defense Access).

What this is not

Not a penetration test. Not an auditor's opinion or a certificate. No promise of zero defects.