Skip to content
Cedric Kato, home
Contact

S-05 — AI agent development · Claude & OpenAI

AI agents that do real work inside your business

Most AI projects stall between a clever demo and something the team relies on every day. I build agents and LLM integrations that plug into your actual systems (inbox, CRM, database, documents) with structured outputs, guardrails and logging, so you can trust them with real work.

Start a conversation What's included

Diagnosis

Sound familiar?

  1. 01

    Your team pastes into ChatGPT all day

    It works, but it's manual, inconsistent and invisible to the rest of the business.

  2. 02

    The pilot never reached production

    The demo impressed, then stalled on data access, reliability and edge cases.

  3. 03

    You can't trust the output

    Answers change from run to run, and there's no way to check what the model did.

  4. 04

    Nobody knows what it costs

    Token spend is a mystery until the invoice arrives.

Scope

AI agents I build

  • S-05.1

    Classification & routing

    Sort emails, tickets, leads or documents and send each one to the right place, the way Sortie does for Gmail.

  • S-05.2

    Extraction & data entry

    Turn PDFs, emails and forms into clean, structured records in your CRM or database.

  • S-05.3

    Drafting assistants

    First drafts of replies, proposals and reports in your voice, with a person approving before anything goes out.

  • S-05.4

    Tool-using agents

    Agents that search, look up records and take actions through APIs, with clear limits on what they're allowed to do.

    See MCP Server Development · S-07

  • S-05.5

    AI features in your product

    LLM features built into your SaaS with evaluation and cost controls. Axis, for example, runs on the Anthropic API.

    See Internal Tools & Dashboards · S-08

  • S-05.6

    Evaluation & monitoring

    Test sets, logging and cost tracking, so you know how well it works and what it costs.

Method

How the engagement runs

I treat an AI agent like any other part of the building: it needs a clear job, defined inputs and outputs, and a way to fail safely. We decide together where the model decides and where a person approves. Then I build it and measure it against real examples from your business.

  1. 01

    Survey

    I map how work actually moves today: every tool, handoff, spreadsheet and inbox. You get the map in writing before anything is built.

  2. 02

    Blueprint

    A scoped plan: what gets automated, what stays human, what it connects to and what it will cost to run. You sign off before I build.

  3. 03

    Build

    Delivered in stages your team can test with real data, with error handling and logging from day one rather than bolted on later.

  4. 04

    Handover

    Documentation and a recorded walkthrough, so your team can run it and change it. Your accounts and your data stay yours.

Note

Step 01 starts with a short note. You get the map in writing before anything is built.

Start a conversation

Evidence

Work I've done & related projects

  1. S-05.N-01

    I built Axis on the Anthropic API, using Claude Code.

  2. S-05.N-02

    For a US client, I designed a central operations dashboard that includes automated Gamma deck generation for prospects.

  3. S-05.N-03

    I helped launch an inbound AI voice agent with GoHighLevel contact capture, sales handoff and SMS fallback.

Sortie landing page: your inbox, finally sorted.
Fig. 01 — AI · Emailgetsortie.xyz

Sortie — visit live site (opens in a new tab)

AI inbox organizer: maps every Gmail sender, classifies them with Claude, then builds the labels and filters so new mail arrives pre-sorted.

  • Claude
  • Gmail API
  • Google sign-in
Axis sign-in screen: the operations OS for service businesses.
Fig. 02 — SaaS · Operationsaxis-ops.app

Axis — visit live site (opens in a new tab)

Multi-tenant SaaS operations platform for VA service businesses — team, clients and communications on one control surface.

  • Next.js
  • Vercel
  • Neon Postgres
  • Resend
  • Anthropic API

Field notes on this

Questions

AI agent questions, answered

Claude or OpenAI: which model do you use?

Whichever fits the task. I build most agents so the model can be swapped, and choose based on quality on your real examples, speed, cost and data policy.

Do you build with Claude Code?

Yes. Claude Code is part of my everyday toolkit, alongside Claude and the Anthropic and OpenAI APIs. Axis, my multi-tenant SaaS, was built with it.

Will our data be used to train the models?

Anthropic and OpenAI don't train on business API data by default. I'll walk you through their data policies and set up the integration so only the data the task needs is sent.

Can an AI agent take actions on its own?

It can, within limits you set. I usually start with the agent proposing actions and a person approving them, then automate the steps that prove reliable.

How do you stop the AI from making things up?

By grounding it in your data, asking for structured outputs, validating them in code and testing against real examples before launch. When it isn't sure, it's designed to say so and escalate.

What does it cost to run?

For internal workflows, usually less than people expect, but it depends on volume and model. I estimate token costs from your real data during scoping and add cost tracking so there are no surprises.

Contact

Tell me what's slowing the business down.

Optional
Optional

I read every message myself.

What happens next

  1. 01I reply personally by email.
  2. 02A short call to walk through your current setup.
  3. 03A written map and a scoped plan, with running costs, before anything is built.
Based in
Metro Manila, PhilippinesGMT+8 · Remote, flexible across US Eastern and AEST hours
Résumé
Download PDF