AI Systems I've Built and Run

A look at the AI systems I have designed, built, and operate day to day. Each one took work that used to run on manual effort and turned it into something that runs on its own or in a fraction of the time. The pages that follow show the systems, the tool-routing logic behind them, and the standards I use to keep AI output reliable.

Anne DeSpain · Park City, Utah
More than 20 years in operations at Amazon, Sundance Catalog, and J. Crew

Anne DeSpain

01 · Proof of Practice

Systems I Run Every Day

These are live workflows, not concepts. Each one took a task that used to eat hours of manual work and turned it into something that runs on its own or in a fraction of the time. Every system here was designed, built, and is maintained end to end.

Automated morning brief

Pulls from email, active projects, and a knowledge base each morning and delivers a finished, prioritized brief as a ready-to-send draft. Replaces a 30-minute manual scan every day.

Automated marketing pipeline

Drafts, schedules, and publishes content across channels on a set cadence, in a consistent brand voice, with a performance dashboard on top. Removes the weekly scramble to produce and post.

Multi-agent research & document workflow

Several AI agents run in parallel to research a topic, pull live web data, and assemble a structured document with sources. Compresses a half-day of research into minutes.

Self-updating knowledge base

After each work session, the system captures what changed and writes it back to a central wiki, so context is never lost between projects or days. Keeps the whole operation running from one source of truth.

Custom systems for specific businesses

Purpose-built tools, including a property-management assistant that handles routine tenant and operations questions. Built to fit how one business actually works, not a generic template.

Anne DeSpain

01 · Proof of Practice

What the Reporting Looks Like

The automated marketing pipeline produces this every day: AI-generated ad creative running on Meta, reviewed by a human, tracked against real numbers. It does more than report metrics, it explains what they mean and what to do next.

A live campaign view from the system. Note the honesty built in: the blended ROAS is flagged as a ceiling, not a floor, and the analysis warns that a high click-through rate did not equal purchase intent. Reporting built to be trusted beats reporting that only looks good. (Company name anonymized for this sample.)

Anne DeSpain

02 · The Core Skill

Matching the Right AI Tool to Each Task

The real skill is knowing which tool wins for which job, then building the workflow around that choice. I stay model- and tool-agnostic. This is the routing logic I use across a marketing and operations practice.

Task	Tool of choice	Why it wins here
Multi-step workflows & system building	Claude	Strongest at agentic, tool-using work and long context. The engine that connects to the actual stack and runs jobs end to end.
Deep research & live web data	Perplexity + Firecrawl	Real-time sources with citations you can verify, plus clean extraction from any page. The right call when the answer must be current and traceable.
Fast content drafting & ideation	ChatGPT	Quick, strong prose for first drafts, subject lines, and brainstorming. The everyday workhorse when speed matters more than depth.
Reporting & summarization	Claude	Handles long documents and transcripts without losing accuracy, so summaries stay faithful to the source.
Image & visual generation	Gemini	Best current results for on-brand images and creative assets, including text rendered inside the image.
Data cleanup & spreadsheet ops	Claude / code	Turns messy exports into clean, structured data by writing and running the logic, not by guessing cell by cell.
Recurring ops (briefs, follow-ups, posting)	Automation layer	Make, n8n, or custom integrations to run the repetitive work unattended and connect the AI tools to the systems.

Anne DeSpain

03 · Repeatable and Enablement-Ready

Sample Workflow SOP

Every workflow I build ships with an SOP and a tested prompt template, so any team member gets the same quality without reinventing the process. Here is a representative example.

Weekly Performance Recap

SOP · Owner: Account Lead · Runtime: ~10 min

Steps

Export the week's channel metrics to the shared sheet (or connect the source directly).
Run the prompt below with the sheet attached.
Review the draft against the flagged numbers, adjust tone if needed.
Paste into the email template and send. Log the recap to the knowledge base.

Prompt template

You are the account lead for [CLIENT]. Using the attached weekly metrics, write a client-ready performance recap. Structure: 1. Headline result in one sentence (the number that matters most). 2. Three wins, each with the metric and the "so what." 3. One area to watch, framed as a next step, not a problem. 4. What we're doing next week. Rules: plain business language, no jargon, no hype. Compare to the prior week where the data allows. Flag any metric that moved more than 20% so a human can sanity-check it before send.

The same pattern (SOP + tested prompt + a human review step on anything unusual) is how I keep AI output reliable enough to put in front of clients and stakeholders.

Anne DeSpain