The AI practice
I do not use AI as a chatbot.I build with it.
A short manifesto and a working toolkit: the principles I default to when a team asks where to start, what to automate, and what to leave alone.
The translation tax is the real problem.
Every project degrades a little at each handoff: stakeholder to sales to PM to design to dev. The first job of AI in an agency is to make that drift visible while it is still cheap to fix.
Start with operations, not output.
The clearest ROI is in the boring, expensive work: briefs, SOWs, status updates, recaps. Start with your most expensive administrative pain, not the most exciting creative experiment.
Automate pattern-matching. Protect judgment.
AI is excellent at competent, derivative work and poor at genuinely original thinking. Automate the pattern-matching ruthlessly so the judgment work gets more of the team’s time, not less.
The output bar is 80% ready to edit.
AI that gets you to a draft you can edit is worth it. AI that gets you to a draft you have to rewrite costs more than it saves. The difference is almost always the inputs.
Structured input, structured output.
Most agency documents already contain the structure someone is about to re-key by hand. Build the automation that turns the doc you already make into the artifact you would otherwise build by hand.
Human-in-the-loop is the default, not the fallback.
AI handles the structured work; a human signs off where reputation lives. That posture is what makes adoption durable: teams trust a tool that does not pretend to be them.
Measure before and after, or do not bother.
Baseline the overhead, deploy, measure again. The delta is the real ROI. Budgets not tied to a specific operational saving are how agencies end up with a stack nobody can defend.
Pattern-recognize, then productize.
A tool I build for one team usually fits four more. The leverage is noticing the pattern early, generalizing the build, and shipping a reusable version before anyone has to ask.
Show, do not tell.
The fastest way to get someone to adopt AI is to hand them a working tool with their own deliverable already in it. Demos before decks, every time.
The goal is creative momentum.
None of this is about AI. It is about time: hours pulled out of data-wrangling and poured back into the work clients actually pay for. Cut the drag, build the leverage, do more of what matters.
What I build
Not slideware. Systems real teams use on a Monday. A few of them:
Project Intelligence Dashboard
A ~350-line Python dashboard wired to Productive.io over JSON:API. Turns time entries and budgets into a weekly burn report.
Cut weekly status reporting from hours to minutes.
Python · JSON:API · Productive.io
SOW-to-Gantt Automation
Reads the phases and hours already sitting in a signed SOW and emits a structured project timeline.
Removes the manual re-keying between sales and delivery.
Python · structured parsing
Kickoff Deck Automation
A 64-slide onboarding deck generated programmatically from a CSV, so every project starts from the same source of truth.
Standardized kickoff across the agency.
Python · pptx
24-Gem ScopeStack Toolkit
A productized library of 24 AI workflows for producers, deployable across a whole agency in one Knowledge Hour.
One build, distributed to every team.
Gemini Gems · ScopeStack
n8n Automation Backbone
A hosted n8n layer that connects the tools an agency already runs into automated, observable workflows.
The plumbing that lets the rest of the systems talk.
n8n · webhooks · MCP
HTX Report
A seven-dataset interactive microsite for Superhuman with a custom GSAP chart system, built end to end with Claude Code.
Research that people actually finish reading.
React · GSAP · Claude Code
The toolkit
The productized version
ScopeStack.ai
24 prebuilt AI workflows that hand agency teams their time back. The tools I kept rebuilding, packaged so the next team gets them on day one.
I embed with teams and leave the system behind.
The best AI deployment is the one nobody talks about six months in, because it has quietly become how the work gets done. If that is what you are after, let us talk.