Pieter VenterEngineer / architect
Start a conversation
Independent systems practiceSouth Africa · remote / periodic travel

Pieter Venter is an architect who ships.

Complex systems.
Clear thinking.

I work where product intent meets operational reality: AI, people, durable workflows and the decisions that make software continue to work after launch.

Explore field studies Start a conversation
15+ years
building software and teams
Early Invisible engineer
architecture through production
Founder
hands-on independent builder
Pieter Venter
Field note / 01Architecture stays close to the code.
Operational trace01 / live record

Intent flows to a decision, then operation, then evidence. Recovery connects decisions to operations when work needs attention.

Read the full working recordRésumé / PDF

Selected systems / 01

Work that has to keep working.

I am drawn to systems with real consequences: unclear handoffs, external failures, human approval, recovery and the need to understand what happened later.

01Invisible Technologies2019–2023

Making complex work flow.

I joined Invisible as engineer #6 and later led its Process Central team. The work centred on a Process Engine that coordinated long-running work across people, software and external systems.

Constraint
Human and software work had to remain observable as it moved through a long-running process.
Decision
Keep the process explicit: a durable orchestration layer, clear handoffs and infrastructure that made operational state inspectable.
Evidence
Led architecture and delivery while managing engineers and engineering managers; stayed hands-on with the critical implementation work.
Read the implementation record
The work

Led architecture and delivery of the Process Engine, alongside technical direction across TypeScript, Postgres, Kafka, Kubernetes and the engineering platform.

The responsibility

Managed engineers and engineering managers, shaped teams and coached technical leaders while staying hands-on with architecture and critical implementation work.

The context

The Process Engine I helped build and lead supported Invisible’s OpenAI work in 2022.

Prisma case study Prisma customer story
02AI Acquisition / Acquisity2025–present

Giving coding agents solid ground.

I build engineering systems around autonomous coding agents: isolated environments, useful logs, CI and review workflows, and safeguards around production changes.

Constraint
Autonomous work still needs a clear owner, useful review points and reliable production boundaries.
Decision
Treat developer experience, QA and deployment controls as product infrastructure rather than cleanup work after the agent runs.
Evidence
Contributed across workflow execution, integrations, permissions, data correctness and production reliability.
Read the implementation record
The work

Contributed across an AI business platform, from workflow execution and integrations to permissions, data correctness and production reliability.

The infrastructure

Improved isolated preview and database environments, structured logging, CI and QA routing, and production deployment safeguards.

The ongoing question

In my own agent tooling, I’m exploring how tasks retain their plan, ownership and approvals when work continues across multiple sessions.

03Fleetnook2026

Agents on their infrastructure. Control in their hands.

I built Fleetnook to explore the handoff between an agent developer and a customer running it: approved Docker deployments, runtime health and signed usage reporting in one control plane.

Constraint
An agent developer needs enough operational signal to help, while the customer stays in control of their environment.
Decision
Use an outbound customer-side runner, customer-approved deployments and revocable, expiring diagnostic access.
Evidence
Built a hosted control plane and customer-side runner with deployment approval, signed telemetry and recovery after connection loss.
Read the implementation record
The build

Built the hosted control plane and customer-side runner, with deployment approvals, signed telemetry, recovery after connection loss and protection against duplicate usage reports.

The boundary

The runner connects outbound from the customer environment. Fixed diagnostics require an expiring grant that the customer can revoke; the customer chooses their own Temporal service.

The state

An independent demo with an invitation-only hosted beta. Interactive support, managed Temporal hosting and payment collection remain future work.

Visit Fleetnook

Additional fieldwork

Syncly2025

Durable data integrations

An independent data integration platform designed around long-running syncs, retries, rate limits, progress reporting and external-system failure boundaries.

Read the architecture study
Mana Finance2023–2025

Research systems that make information useful

AI-powered portfolio monitoring, SEC filing analysis and scheduled research workflows for downstream analysis.

Independent build
BabyTracker2026

A small app for busy days

A native iPhone app for logging feeds, sleep and care, built with SwiftUI, SwiftData, widgets and App Intents.

Independent build

Live field studies / 02

Show the working idea.

Interactive concept demos built to make a workflow tangible before a full product exists. They use fictional data and simulated actions; no client data, live inbox or production integration is shown.

VERN Signal Desk prototype showing a sales follow-up review workspace.01

Sales follow-up prototype

Signal Desk

A review-first follow-up desk for a spreadsheet-led sales process. It drafts within supplied context, keeps active conversations with people and makes every simulated decision traceable.

Open the demo Read the study
Operations Circuit workflow study showing a dark control surface.02

Embedded workflow concept

Operations Circuit

A bounded workflow builder for product teams: show the route, make approvals explicit and keep recovery work visible.

Open the demo Read the study
Order Bench wholesale order-intake study showing a paper-like review surface.03

Wholesale intake concept

Order Bench

A human review surface that turns an emailed wholesale order into a structured draft without pretending every ambiguity is resolved.

Open the demo Read the study
View all field studies

Public record / 03

The work leaves a trail.

A small selection of source material for people who prefer to inspect the details.

tRPC / 2021

Early contributor and maintainer.

Collaborated with Alex Johansson on early API design and reviewed middleware, server-side calling APIs and the v9 changes.

Prisma / 2021

Scaling a system without stopping it.

Prisma’s case study documents a gradual Sequelize-to-Prisma migration at Invisible, with no customer downtime reported during the transition.

Open source / ongoing

Small changes, real failure modes.

Documentation credit in ts-rest and a Factory transition-handling fix with regression coverage in Mastra.

Working style / 04

Close to the code. Clear on the outcome.

01

Make the first useful boundary small.

Find the smallest version that a real person can use, learn from it and let that shape the architecture.

02

Keep ownership and review visible.

Teams move faster when decisions, approvals and handoffs are explicit enough to inspect later.

03

Design for recovery.

External systems fail. People change their minds. Good software makes the next safe action obvious.

I’m at my best in small teams with hard problems and short paths from idea to reality. I’ve led engineers and engineering managers, but I’ve never wanted to get too far from the work itself.

TypeScript, React, Node.js, Python and Postgres are familiar ground. Durable workflows, integrations and infrastructure for AI agents are the problems I keep coming back to.

Start a conversation / 05

A worthwhile problem is a good place to begin.

I’m interested in small, remote teams building something they believe in, with room to shape the architecture, own the outcome and keep shipping.

Book a call Read my résumé
GitHub LinkedIn
Remote · open to periodic travel© 2026 Pieter Venter