Infrastructure and platform engineering for funded startups

Built once. Yours to own.

We build your startup's core infrastructure and deployment pipeline in one to four months, with security built in from the first commit, for a team where AI writes much of the code. Then we train your engineers to run it, and step back.

The plan is free, with no strings attached: yours to build with your own team and your AI.

AGENTSsandboxedscoped · loggedworkingDEVacct: devSTAGINGacct: stagingPRODacct: prodttl=1h · admin: 0CI/CDpolicy gatesSECRETSnone in reposDATAread by role
Plan view · one buildEvery person and agent holds one compartment
1–4 monthsto a working foundation
Yoursevery repo and account stays with your team
By projectpriced by milestone, never by the hour
Any cloudAWS, GCP or Azure
Who it's for

Funded startups with nobody senior in infrastructure.

  • Just raisedPre-seed, seed or Series A, with the budget to build it right
  • 5–30 engineersShipping fast, and nobody owns infrastructure full time
  • Money, health or personal dataWhere one mistake costs the most
  • Building with AIAgents writing code, opening PRs, touching systems
Why now

Access grows faster than anyone reviews it.

Skip this part
Without bulkheads

One key reaches everything.

One shared account, broad roles and keys that never expire. A single leaked key floods the whole ship: production, data, secrets and your agents with it.

With bulkheads

The same leak stays in staging.

Staging has its own account and its own credentials, so the key reaches nothing else. Production, data and your agents keep working.

Bulkhead (n.): the wall that splits a hull into watertight compartments, so one breach doesn't sink the ship.

Rotate and rebuild

Rebuilt from code.

The key is rotated and staging is rebuilt from code. Nothing was made by hand, so nothing is lost.

CIPRODDATASTAGINGAGENTSSECRETS
What we build · AWS, GCP or Azure

The standard build, layer by layer.

Skip this part
Layer 1 · Foundation

Accounts, network, identity.

Separate production, staging and development accounts, the network, single sign-on, roles that match jobs, and a secrets manager. Least privilege is enforced as code from the first commit.

Cloud accountsNetworkSSO and rolesSecretsInfrastructure as code
Layer 2 · Delivery

One path to production.

CI/CD with review gates and policy checks. Adding a service is easy, and an unsafe change stops before it merges. Agents get isolated sandboxes with scoped, logged access.

CI/CDPolicy checksAgent sandboxesSupply-chain basics
Layer 3 · Runtime and data

Where the product runs.

Managed containers, or Kubernetes when you actually need it. Managed databases with tested restores, and data pipelines where access is decided by role. You own the data.

ComputeDatabasesData pipelines
Layer 4 · Application

Wired and watched.

Configuration, service-to-service auth and permissioned APIs behind a gateway. Logs, metrics and alerts your team owns and actually watches.

App wiringPermissioned APIsObservability
Layer 5 · Your team

The people who run it.

Runbooks for everything we built, and training for two owners on your team. The top of the build is the people, not more tools.

RunbooksTrainingTwo owners
Foundationaccounts · network · identity · secrets · IaCDeliveryCI/CD · policy checks · agent sandboxesRuntime and datacompute · databases · pipelinesApplicationwiring · APIs · observabilityYour teamrunbooks · training

Referred outOn-call and incident response, penetration testing, compliance programs and company IT. We point you to people who do them well.

For AI-driven teams

Every change takes the same path.

Skip this part
Change opened

A PR, from a person or an agent.

People and agents ship through one pipeline. Infrastructure is code, so every change is a diff someone can read.

Policy check

Rules run first.

Checks run on the plan before anyone reads it. A change that widens production access stops here and gets narrowed in the same PR.

Review and merge

A person approves, on the record.

An owner on your team reads it and signs off. The merge records who asked for the change and why.

Staging, then production

Applied by the pipeline.

A short-lived role applies it to staging, then promotes it. No person and no agent holds standing admin in production.

  1. 1PR
  2. 2Policy
  3. 3Review
  4. 4Merge
  5. 5Staging
  6. 6Prod

$ pr #142 opened · add a payments bucket

✗ policy/prod-access · wildcard action on prod · stopped

✓ narrowed to the one action it needs · passes

✓ review · approved by an owner on your team

✓ merged · recorded with who asked for it

→ staging · role/deploy-staging · ttl=1h

→ prod · role/deploy-prod · ttl=1h · humans=0

Every agent gets

  • Its own sandbox, rebuilt from code
  • Access scoped to the job, short-lived
  • Every call logged, in your systems
  • Infrastructure changes only through PRs

Nobody gets

  • Standing admin in production
  • Wildcard permissions
  • Shared or long-lived keys
  • Customer data without a role that needs it
Instead of the hire

The foundation is a project, not a job.

Skip this part
The usual answer

A senior hire, year one.

In New York: about $220K base, about $95K in benefits and payroll, and about $55K to an agency. Roughly $370K, plus equity, and they start 47 to 95 days after you begin looking.

The build

$50K to $200K, once.

A serious investment, scoped to your team and paid by milestone. One to four months, and most of it is done before a new hire would have ramped up.

Year two

The hire costs it again.

About $315K a year from then on, for a senior role with less senior work left. The build costs nothing more, unless you choose a retainer.

Hire · year one
$370K
Build · once
$50–200K
Hire · year two
$315K
Build · year two
$0
BaseBenefitsAgencyBuild from $50Kup to $200K

Hire: $220K base, benefits and payroll at 43% of wages (BLS, June 2026), a 25% agency fee. Salary sources disagree and come mostly from vendors. Checked 2026-10-17.

Do you need the hire at all?Get a free plan
How it works

Built to hand over, not to rent.

Skip this part
Free · no strings attached · about 3 weeks

A free plan

We look at what you have and write a plan for your coding agent to build out. It's yours to keep and run yourselves, with your AI, at your own pace. No catch, and no obligation to hire us.

1–4 months · by milestone

Build

We build it as PRs into your repositories, and your engineers review and merge every one. Most of it lands in the first two months.

Part of the build

Training

Two of your engineers become owners. They run every runbook with us before we step back.

The keys move to you

Handover

Everything is code in your repos and runs in your accounts. We keep no standing access, no data and no secrets.

# your-company/infraaccounts/ identity/ policies/ pipelines/ services/ runbooks/ OWNERS two names from your team
Yours, at your pace

After

You run it and grow it. Help after handover is included. Bring us back for a review, a retainer or an advisory role, if and when you want.

BULKHEAD LABSYOUR TEAMCONSULTBUILDTRAINHANDOVERYOURSBar length: share of the work at each stage. The orange square is the keys.
Who builds it

Abdul Rabbani

An infrastructure and security engineer who has spent years on early teams, and the person who designs and builds your foundation.

How we think about security

  • Years inside startups

    Early teams in trading, crypto and data, where speed, money and reliability all mattered at once.

  • Built from the ground up

    Foundations built from an empty cloud account, more than once, then handed to the team that runs them.

  • Fast, without the shortcuts

    Least privilege and isolation from the first commit are what keep a startup fast later.

  • By startup engineers

    For startup engineers: MVP first, then iterate, delivered milestone by milestone.

Get in touch

Tell us what you're building.

Your email and a few words are enough. You get a written plan, free and with no strings attached, that your team can build yourselves with your AI.

  1. 1

    A reply

    Within two business days, from the engineer who would do the work.

  2. 2

    A first call

    Your stack, your team, and what worries you. No paperwork.

  3. 3

    A written plan

    Within about three weeks, written for your agent to build from.

Or write to hello@bulkheadlabs.com

Add context optional

In person in New York City · remote anywhere in the US