WORLD/01.STUDIO
● ONLINE
PHT
10°43′N · 122°34′E
v3.0 — QUIETLY.BUILT.LOUDLY.SHIPPED
SCROLL
000%
ASTHER · LOUIE · CABARDO · 2026
FULL—STACK · ENGINEER · PH
← back to services
Service · 04

Agentic Engineering.

Production-grade integration of AI into your product or workflow. Claude as a peer, not a chat box — with proper context budgets, prompt caching, and agent loops that actually finish their tasks.

// agent loopBoundaries first. Tools second. Cost visible.
  1. 01
    User input
    Slack message · web form · email
  2. 02
    Context fetch
    MCP · cached prompts · KB search
  3. 03
    Plan
    Sonnet 4.6 · tool catalog
  4. 04
    Tool loop
    exec → observe → reflect
  5. 05
    Schema verify
    structured output · zod
  6. 06
    Action / Reply
    DB write · email · ticket
Observability
  • tokens / task2.1k
  • p50 latency1.4s
  • fail-rate1.8%
  • $ / 1k tasks$0.42
// where AI helps

Six shapes I've shipped — pick the one that matches your problem.

Not every workflow is an agent. The trick is picking the shape that fits, then locking the boundaries before writing a prompt.

internal · public

Customer support copilot

Reads conversation context, drafts responses, routes the easy ones, escalates the hard.

models
  • Sonnet 4.6
  • Haiku 4.5
tools
  • MCP · CRM
  • MCP · KB
  • Email send
  • Ticket create
internal

Internal-ops agent

Replaces a 10-person manual workflow — runs at midnight, queues exceptions for review.

models
  • Sonnet 4.6
tools
  • Postgres
  • Slack
  • Sheets
  • MCP · Notion
engineering

Code review reviewer

Reviews PRs against a project skill (style, conventions, security) — comments inline.

models
  • Sonnet 4.6
tools
  • GitHub MCP
  • Diff parser
  • Skill engine
data

Research / scrape agent

Background agent that pulls from public sources, normalizes, and writes to your DB.

models
  • Haiku 4.5
tools
  • Browser MCP
  • Postgres
  • Schema validator
public

On-product copilot (chat)

An in-product chat that knows the user's data and the actions they can take.

models
  • Sonnet 4.6
  • Opus 4.7
tools
  • Tool use
  • Streaming
  • Structured output
product

Workflow / agent platform

If your product IS the agent — Flowys-style. We build the runtime + the editor + the deploy.

models
  • all
tools
  • Tool registry
  • Sandbox
  • Logs
  • Replay
// pick the model that matches the task
Opus 4.7
— heavyweight

architecture, hard reasoning, tricky debugging

Sonnet 4.6
— workhorse

everyday agent loops, code review, copilots

Haiku 4.5
— sprinter

transformations, classification, background work

// included

What's included.

  • +Use-case shaping (where AI helps, where it just adds latency)
  • +Model selection (Opus / Sonnet / Haiku) and cost forecasting
  • +Prompt caching, structured outputs, and tool-use loops
  • +Skills-based agent design (long-lived, reusable specifications)
  • +MCP server integration for internal tools
  • +Observability: tokens, cost, fail-rate, escape hatches
// outcomes

What you can expect.

  • ▶An AI feature you can defend in a board meeting — not a demo gimmick
  • ▶Cost-per-task you actually understand and can budget against
  • ▶An agent that gets better month-over-month because the context budget is managed
// process

How an engagement runs.

01
Frame

We pick one workflow, draw the data flow on a whiteboard, and decide what the agent should NOT do. The skill is in the boundaries.

02
Prototype

Smallest possible loop. Real tools, real data, real failures. We watch token cost and edit the prompt with intent.

03
Harden

Caching, retries, schema-validated outputs, fallback to a smaller model, observability, and a kill switch.

04
Ship

Production deploy with metrics dashboards. Cost-per-task and fail-rate visible from day one.

// best fit

Best for.

  • ◇Teams whose first AI feature shipped and now they want it not to embarrass them
  • ◇Founders building an AI-native product (chat, agent, copilot) and want it solid
  • ◇Internal tooling teams replacing a 10-person manual workflow with an agent
// pricing

Engagement & pricing.

AI feature integration: 4–8 weeks. AI-first product: scoped as full Product Engineering.

Default stack
  • Claude API (Opus/Sonnet/Haiku)
  • Vercel AI Gateway
  • AI SDK v6
  • MCP servers
  • Structured outputs
  • Tracing
// related

Recent projects in this lane.

Workflow AutomationFlowysA personal workflow automation app — design, test, and run flows in one place, with AI steps and webhook triggers built in.↗

Ready to start?
Send a one-paragraph brief.

What you're building, the rough timeline, and one constraint that matters. I'll reply within a day with a one-page response and a quote.

Email a brief →See the archive →