Andres Espinoza · multidisciplinary operator and builder
AI won’t replace people.People who use it well will outpace those who don’t.
My point of view
Technology matters when it helps people use what they know, make better decisions, and spend less time fighting the process.
The person behind the work.
I have always learned by building, competing, and bringing people together around difficult things.
The through-line
The titles changed. The work underneath them stayed surprisingly consistent: understand what people need, connect the right disciplines, and carry the idea through execution.
Project delivery
In construction and engineering, I coordinated vendors, deadlines, clients, project teams, and company leadership. I learned to translate across people who measured success differently.
Creative direction
At an advertising agency, I led work from discovery and concept through production. The idea had to be good, commercially useful, on time, and executable by the team.
Business operations
At a food and beverage manufacturer, my role grew from design into marketing leadership across sales, retail, packaging, product development, e-commerce, and operations.
Real-world ownership
Running service, drone, rental, and software work taught me the part job descriptions miss: customers still need an answer, tradeoffs cost money, and the result has to work after launch.
How I work
I learned this approach before generative AI existed. The newer tools let me move farther into implementation, but the method still begins with the people and the work.
Live the work
I go to the people using the process. Their experience matters more than my first assumption.
Find the friction
I separate the visible complaint from the decision, handoff, or repeated task underneath it.
Connect the parts
I define who knows what, where decisions belong, and what the process or product needs to make those decisions easier.
Make it testable
I choose the smallest useful response. It may be a process change, a normal software rule, an AI-assisted system, or a mix.
Compare and decide
I compare the new way with the old one. Usefulness, adoption, safety, and policy tell me to keep, change, or stop.
Three recent examples.
Not the whole story.
These projects show what changed when AI and agent-directed development expanded the range of solutions I could personally direct. The thinking behind them came from the career above.
Artwork review
Three accountable approvals
What I saw
I followed the review through five teams. The same comparisons happened again and again, while status and ownership were hard to see.
What I decided
Use AI and comparison tools for preparation and repeat checks. Keep Technical Services, Marketing, and Sales responsible for approval.
How I judged it
Compare review time and issue quality with the prior process, while treating enterprise policy and user trust as part of whether the system works.
Learn the storyFive teams → bounded AI work → three accountable sign-offs
Every revision restarted the room.
Five teams repeated broad comparisons even when only a small area changed. The work was high-stakes, but scarce experts kept returning to the same file.
Move repetition—not authority—to AI.
AI performs defined first-pass and revision comparisons. Technical Services approves specifications, Marketing approves artwork, and Sales approves customer and product alignment.
Faster review with people still responsible.
Process-owner estimates put first review at 30–45 minutes before and no more than 15 after; later revisions moved from 20–25 minutes to 2–3. These are observations, not a formal time study.
Workflow and policy before software.
I owned the process, built the initial prototype, pushed the AI-policy question, shaped the operating boundary, and helped deploy the governed workflow. Implementation was AI-assisted and cross-functional.
Decision architecture / anonymized
COACH
before prescription
What I saw
A standard training plan had structure, but it could not respond when sleep, recovery, soreness, schedule, or completed work changed.
What I decided
Deterministic code owns readiness and safety limits. AI explains, drafts, and converses. The athlete can approve, adjust, or veto consequential changes.
How I judged it
I use it daily. A useful decision has to reflect current data, explain itself, reach the watch, and remain easy to change. I do not claim proven performance gains.
Learn the storyA flexible week becomes today’s specific workout
Sunday’s plan can be wrong by Wednesday.
Recovery and schedule change after a weekly plan is written. The athlete still needs a concrete decision that morning—not another dashboard score.
Wait, reconcile, then prescribe.
The system waits for two overnight wearable feeds, normalizes them into one deterministic readiness state, and materializes the day’s workout inside the larger training block.
The athlete remains the final authority.
AI can explain and converse, but deterministic logic owns the training boundary. The athlete can approve, adjust, or veto before a workout is sent to the training device.
Used daily; measured without clinical claims.
Across 37 days, 33 of 36 planned days materialized, all 31 non-rest workouts had recorded device delivery, 10 adjustments were applied, and 21 debriefs ran. AI agents wrote the code under my direction.
abodly
included stopping
What I saw
My wife, our realtor, and I were making one decision across listing sites, texts, showing days, memory, and different priorities.
What I decided
Build one shared workspace that compared homes around the life we would live, not only bedrooms, price, and listing photos.
How I judged it
The tool helped our own search. The public path did not survive the listing-data and compliance review, so I stopped commercialization and kept it internal.
Learn the storyA household decision—not another listing portal
One move lived in too many places.
Listings, rankings, commute tradeoffs, tour notes, and conversations were fragmented across sites, screenshots, texts, and memory.
Put the shared decision in one workspace.
Abodly combines shared boards, rankings, notes, commute comparisons, tour planning, and synchronized tour mode for the household and realtor.
Stopping was part of the product decision.
My wife, our realtor, and I used the system. I kept it internal when the broader listing-data model crossed my compliance and platform-risk boundary.
Founder and operator—not sole code author.
I defined the problem, users, priorities, risk posture, requirements, and acceptance criteria; directed AI agents that implemented and tested it; decided what shipped; and operated the product.
Personal agent system / live operating model
The system behind
the systems.
I did not want another chatbot window. I built an operating model that can meet me in a message, carry context across tools, route work to the right agent, and stop when the decision belongs to me.
The point is continuity with boundaries: shared context where it helps, separate credentials and sensitive domains where it matters, and explicit human approval before consequential action.
GitHub / verified snapshot
Build activity.
Private by design.
The work is active even when the repositories are not public. This view shows authenticated GitHub contribution totals without exposing private clients, employers, code, or repository names.
past 90 days
past 365 days
past 90 days
1195 private2 public
How the work gets built.
I conceived, directed, tested, and operated these systems. AI agents wrote much of the code under my direction and review.
I am not presenting myself as the engineer who typed every line. I own the problem, product direction, requirements, tradeoffs, testing, acceptance, and whether the result is useful enough to keep.