Skip to content
wojciech.io

AI Systems

AI in the architecture. Not bolted on at the end.

Most teams add AI to tasks. I add it to systems. The difference is whether you get a faster copywriter or a different operating model: one that compounds, self-corrects, and runs without a human in every loop.

gtm-agent · claude-code

$ claude run gtm-agent --client acme-saas

▸ Loading CLAUDE.md operator identity...

▸ Connecting MCP tools (Ahrefs, HubSpot, LinkedIn)...

▸ Reading client memory: acme-saas.md

▸ Fetching ICP signal for 247 prospects...

▸ Scoring leads against ICP matrix...

12 high-fit leads enriched

Sequences drafted and staged in HubSpot

Memory file updated with new signals

Cost: $3.40 · Time: 4m 12s

$ _

How I think about AI

Four principles.

01

Fix the system first

AI doesn't fix a broken GTM. If the pipeline leaks at qualification, AI-personalized sequences still go nowhere. I design the system first, then find where AI creates real leverage.

02

Operator-grade, not demo-grade

Generic AI wrappers don't survive contact with real work. I build tools that are instrumented, specific to the workflow, and used in production, not shown in a slide deck.

03

Compounding, not one-shots

The goal is a workflow that gets faster and smarter over time. Not a prompt that impresses once, breaks on the second run, and gets abandoned after two weeks.

04

Every loop is measured

AI workflows without feedback loops don't improve. They drift. Every system I build has measurement baked in, so you know what's working, what's not, and what to change.

The operating loop

Where AI sits in the system.

Not a task assistant bolted on at the end. AI runs the enrichment, scoring and execution; the operator owns the decision; every loop is measured and feeds the next.

Signals GA4 · intent · CRM AI Enrich firmo · tech · people AI Score · ICP 0–100 fit HUMAN Decide operator owns this AI Act outbound · content · ads Measure dashboards · D1 Compounding loop: every cycle is measured and feeds the next

Tech stack

Tools I actually use.

Reasoning

Claude OpusGPT-4oGemini

Coding & agents

Claude CodeCursorCodex

Automation

Maken8nZapier

Analytics

AmplitudeMixpanelLooker

CRM

HubSpotSalesforcePipedrive

Enrichment

AhrefsBrandwatchSimilarWeb

What I've built

AI workflows and tools in production.

Ads · LLMs · GTM

AdsAI / Ad Assistant

One full ad lifecycle in AI: brief → generate → score → iterate → prepare variants for deployment. Built for GTM operators, not agency workflows.

Claude Code · Agents · Open source

Claude Code GTM Agent Starter Pack

Open-source foundation for building GTM agents with Claude Code. Designed so operators can ship without starting from scratch.

View

macOS · Swift · Codex

Notch, native macOS app

Built with Codex and Xcode. Proof that AI-assisted development covers native apps, not just web, extending the operator's reach beyond the browser.

View

AI adoption · Framework · B2B SaaS

AI Adoption playbook for B2B SaaS

Internal framework for compounding AI adoption across a B2B SaaS company: from growth and marketing through to product and ops.

CRM · Outbound · AI

CRM + outbound AI workflows

Prospecting, sequencing, scoring, and CRM enrichment, redesigned with AI in the loop to cut manual work and improve signal quality.

SEO · Content · AI

SEO + content AI workflows

From keyword research and brief generation to programmatic content scoring. Built to scale content production without scaling headcount.

Live demo

A shipped AI tool.

GrowthHub: a B2B SaaS growth dashboard with live funnel metrics, pipeline snapshots and ICP segment tracking — synced nightly from Pipedrive.

Open live demo No login required · synthetic data · built on Cloudflare D1 + Pipedrive

Work with me

If you want AI in your GTM stack, done right.

Not a workshop. Not a deck of use cases. An actual system: built, shipped, and measured.

AI GTM audit: where does AI actually help vs. add noise

Workflow redesign with AI in the critical path

Custom agent and tool development

Team onboarding to AI-native operating patterns

Ongoing iteration and measurement

Next step

Want AI in your GTM stack, done right?

Not a workshop. Not a use-case deck. An actual system: built, instrumented, and running. Book a call.