Fractional CTO

MVP triage

From vibe coding to production.

Your AI-built MVP worked. Now it has to survive production. I audit, stabilize, and get to production apps built with Lovable, Bolt, Replit, Cursor, or Claude Code, keeping what works instead of rewriting everything.

Book a Free 30-min Call

You leave knowing your biggest production risk.

The starting point

Built with AI tools? That was the right call.

Lovable, Bolt, Replit, Cursor, Claude Code. If your MVP came out of one of these, you already did the hard part: you turned an idea into something people can actually try. Maybe it has paying customers, maybe nobody outside your circle has touched it yet. Either way, you're holding a working product instead of a slide deck. What these tools don't give you is what production demands. Isolation between customers, secrets that stay secret, tests that catch regressions, and a way to roll back a bad release.

Lovable Bolt Replit Cursor Claude Code

Diagnosis

The patterns I keep finding

These come from real calls with founders and from audits I've run:

One customer can see another's data

Multi-tenant isolation living in the app layer, or bypassed entirely because a workflow tool connects with an admin key. Nobody has tested that one customer can't read another's data.

The AI answers from memory

A RAG product with no source citations and no evaluation harness. You find out quality dropped when a customer tells you.

Workflows nobody can review

Core business logic living in JSON blobs inside a low-code tool: no version control, no code review, no tests.

Secrets in the open

API keys in client code, service credentials shared across environments, no rotation story.

Payments held together with hope

Webhooks without retries or idempotency. One missed event and your books don't match your provider's.

No way back

Deploys straight to production with no rollback path. Every release is a bet you can't unwind.

Agents that ship straight to production

Coding agents with push access publish to main, and every push deploys. Nobody reviews the change, and the rules meant to stop them live in files the same agent can edit. I start with a guardrail that holds back only changes to sensitive paths, then move it into branch protection. See the case study ↓

Rules that only exist in a prompt

Your agent's instructions say "every endpoint checks the user's role" and "never commit customer data." To the agent, that's a suggestion: it weighs it against finishing the task, writes the tests that check its own work, and can edit the rules file. I turn each rule into a check that fails the build (a test that calls every endpoint as the wrong user, a scanner that blocks customer data at commit) and put those checks where the agent can't change them. See the case study ↓

How it works

How I work

Start with the audit. Everything after it is optional.

1

Production Readiness Audit

$4K / 2–3 weeks

Written findings across security, tenancy, payments, tests, and deploys. A keep/fix/rewrite decision for every component. A prioritized 30-day plan, and in the first week, a list of what to close right away.

2

Vibe-to-Production cycle

4–8 weeks · scoped by the audit

Hands-on fixes and build-out to a production beta. The audit defines the scope, so you pay for known work instead of padded estimates.

3

Continuity, two ways

$2K /mo care plan

A care plan ($2K/mo: monitoring, dependency updates, small fixes) or a fractional CTO retainer (from $5K/mo, same tiers as my pricing page). Either way you also get the handoff kit: documentation, an AI-native delivery loop, and help hiring your first engineer if you'd rather own it.

The decision

Keep, fix, or rewrite: decided per component

"Rewrite it properly" is the most expensive sentence in software. The audit ends with an explicit decision for each part of your system: keep (it works and the risk is acceptable), fix (targeted repairs, days not months), or rewrite (only where the component blocks the business). Most vibe-coded MVPs land mostly in the first two columns.

Keep

It works and the risk is acceptable.

Fix

Targeted repairs, days not months.

Rewrite

Only where the component blocks the business.

In practice

An agent-built platform, already in production

Audit & containment · Wealth management

Wealth management platform

A regulated wealth management firm built its internal platform with a team of non-developers and Claude Code agents. They use it every day for client onboarding files, CRM, advisor billing, AUM tracking and the order desk. It runs on Supabase and Netlify. I audited it, then worked inside their team to close the most urgent issues.

148
findings, 2 of them critical
2 weeks
of containment, with their team

See other assessments I have run

  • I spent three weeks on security, compliance and architecture, and came back with 148 findings (2 critical, 21 high) and 59 things worth keeping.
  • A second AI model challenged every finding, and I checked each objection against the code. With an agent running on an open-weights model, I reproduced the riskiest authorization and security findings in an isolated local environment with synthetic data.
  • Before any AI read the repository, I redacted it with patterns for IDs, tax numbers, bank accounts and emails, a secrets scanner, and a NER model running offline for people's names.
  • In two weeks of containment, their team and I enforced two-factor authentication in the API and in the database, stopped agents from publishing to sensitive paths without review, replaced broad personal tokens with a scoped GitHub App, and rotated credentials.
  • I wrote architecture decision records and per-area review criteria that the AI reviewer now reads before anything is published. I added tests that fail when a change breaks an architectural rule, and a data model diagram generated from the migrations. The rest went into a roadmap of 40 initiatives, each with a testable exit criterion.

Background

About Me

I'm Ezequiel Actis Grosso, a technology leader with 25 years in software, from Buenos Aires to Miami, and from San Francisco to Sydney, across global corporations and fast-growing startups.

For the past decade I've worked with founders in Latin America and the United States, helping them move fast, find product-market fit and scale their B2B SaaS products, without the over-engineering that slows them down.

Ezequiel Actis Grosso

Ezequiel Actis Grosso

Fractional CTO

Lean Studio · LinkedIn

Next step

Ready to find out where your MVP stands?

A free consultation. You leave knowing your biggest production risk.

Book a Free 30-min Call

FAQ

Questions founders ask

My app is live and working. Why wouldn't it survive production?

Working for 20 friendly users and working under real load, real attackers, and real customer data are different claims. The audit tests the second one.

Will you rewrite everything?

No. The default assumption is that your prototype is an asset. Every component gets an explicit keep/fix/rewrite call, and most components don't need a rewrite.

Do I need to hire developers after you finish?

No. You choose: a care plan keeps the app maintained ($2K/mo), I stay on as your fractional CTO (retainers from $5K/mo), or you take the handoff kit (documentation, the delivery loop, and help hiring your first engineer) and own it.

Can you give me a fixed quote for the whole thing up front?

For the audit, yes: $4K, two to three weeks. The build cycle I quote after the audit: pricing unknown work up front just means padding it.

What do you need from me to start?

Access to the code wherever it lives (a Git repo, or the project inside Lovable, Replit, or Bolt), a walkthrough of the product, and 30 minutes. NDAs are fine.