Bruno Saldivar

Services · Agentic engineering

Your team already uses AI coding agents. I build the workflow around them.

Cursor, Claude Code, Codex: the tools are good. What is usually missing is the structure around them. Who reviewed this change? Was it tested? What did the agent touch, and what did it cost? I design and build that structure on your own repository, CI and AI accounts.

Start with a diagnostic See the method

The problem, and the fix

Same tools. A workflow you can trust.

Before

Before: each person uses an agent their own way and changes go straight to main with no common test, review or record. Cursorown habits Claude Codeown prompts Codexown rules tested? reviewed? cost? what changed? main
Before. Everyone uses the agents their own way. Nobody can say what changed, whether it was tested, who reviewed it, or what it cost.

After

After: a scoped task runs in an isolated workspace and branch, passes a test gate against main, gets an independent review, a human approval where agreed, and merges with a run log and alerts. Task, scoped Isolated workspace + branch Test gate: not worse than main Independent review Human approval, where agreed Merge + run log + alerts
After. Same tools, one workflow: every task isolated and gated by tests, reviewed independently on the routes you agree, approved by a person where it matters, and logged.

How an engagement works

Three steps, each with a clear end.

  1. Step 1

    Diagnostic

    Fixed scope

    A short, read-only look at how your team works with AI agents today, and a concrete recommendation for your case. Low risk: you decide afterwards whether to go further.

    • How each person uses agents today
    • Risks: secrets, permissions, missing tests, who reviews what
    • Which tasks an agent can do alone, with review, or never
    • A workspace design for your repository and tools
    • A plan for the first month
    • You get: a written report and a readout call

    Limits. Read-only access to one repository. Your code is not changed during the diagnostic.

  2. Step 2

    Custom implementation

    On your tools

    The workflow, built into your own GitHub or GitLab, CI and agent tools, then run end to end on real tasks with your team.

    • Task isolation: one workspace and branch per task
    • Agent policies: instructions, allowed files, no-go zones
    • Test gates: nothing merges if the suite gets worse than main
    • Independent review on the routes you agree
    • Human approval points where they matter, such as payments or customer data
    • Metrics, logs and alerts: first-pass approval, rework, review rounds, tokens where the provider reports them
    • Runbook and handover, so your team runs it without me

    Limits. Agents get no production access. A test gate protects only what your tests cover; I add minimal smoke tests where the gate needs them, not a full test suite.

  3. Step 3

    Maintenance and improvements

    Optional

    Updates, tuning from your own metrics, new integrations and fixes, in the format that suits you.

    Hour bank

    A block of hours agreed in advance, used as you need it and reported against the work done.

    Per intervention

    One request at a time, scoped and quoted before any work starts.

    Clear lines. Defects in what I delivered are fixed under the warranty we agree, not billed as maintenance. New scope is quoted separately. Support runs on business days; there is no 24/7 on-call.

Case study · proprietary platform

AWOP · Agentic Workflow Orchestration Platform

Proprietary platform · built end to end · since September 2026

Multi-agent orchestration and coordination: agents share task ownership, handoffs and review rules. MCP (Model Context Protocol) connects assistants to the workflow; evals measure runs by model and role. Self-healing mechanisms handle stalled work and failed handoffs within explicit limits.

Several AI agents work in parallel under one shared engineering standard. It is where the method I offer was built and where it is applied: the trading decision engine and the execution copilot are developed with it, and the energy research platform was. AI agents ship code, independent agents review it and tests decide what merges, within explicit limits rather than unbounded automation.

Its limits, stated up front: agents work in isolated branches and working copies, which is not a security sandbox; and independent review is mandatory on autopilot merges (changes that merge without a person, on routes agreed in advance), not on every merge.

Full background →

341tasks merged across the four projects by AI agents, every one through the test gate.

A test-regression gate on every merge: nothing merges if the tests get worse than main. Measured per run: first-pass approval, rework rate, review rounds to acceptance, tokens per accepted task. Task count measured on 20 October 2026.

Multi-agent coordination, self-healing and measured evidence →

Ownership

You keep everything. Nothing to license.

Next step

Start with a diagnostic.

Book a 30-minute intro call. We look at how your team uses agents today and whether a diagnostic makes sense. The diagnostic is quoted at a fixed price before you commit; any implementation is quoted from what the diagnostic finds. Remote, in English or Spanish.

Book a 30-min intro bru.daytrader@gmail.com LinkedIn

Not offered: security certifications, 24/7 support, trading signals or investment advice.