Build Things That Matter

At fourteen I learned software the hard way: cheap shared hosting, hand-editing httpd.conf, breaking a site with one rewrite rule and staying up to fix it. Constraints were my first teachers. If I didn’t understand how something worked, it didn’t work.

Then AWS turned infrastructure into a menu. I didn’t stop caring about the plumbing—I just spent more time on the product. We are at that kind of moment again. Only now the menu is automation. AI can sketch a system, draft code, and wire the basics while you are still talking to the user.

People ask if AI replaces developers. Wrong question. In a good shop, AI is an apprentice. It sets the jigs and speeds the cut. We choose what to build and where the bar is.

At Voortgang our rule is simple: build things that matter. That means removing real pain with narrow software, then stacking those wins until they look like a platform. No theater, no screenshots for applause. Usefulness people would miss if you took it away.

CareHub (tenant management without chaos)

We built a simple system for tenants and property teams—announcements, ticketing, a living knowledge base, and a clean escalation path. AI helps categorize issues, suggest responses, and surface duplicates so the team fixes once instead of ten times. Result: fewer WhatsApp fire drills and faster, cleaner resolutions.

Global digital awards voting (credibility first)

A voting system that handles nominations, identity checks, fraud detection, and a real-time, auditable tally. AI clusters suspicious patterns and flags anomalies for human review. What used to be a tangle of forms and spreadsheets became one trusted source of truth.

Marketing Portal (alignment that pays for itself)

One place for creative requests, C‑Suite visibility, and performance marketing. AI summarizes weekly changes, highlights outliers in spend and CPL, and generates tidy rollups for leadership. It helped drive a pipeline that converted $2M in sales in two months on $30k ad spend—not because it was flashy, but because everyone made decisions from the same sheet of music.

AI did not “do” these projects. It let us shorten the loop: define the job, cut a walking path, ship a thin slice, learn, repeat. The scarce resource was not model quality; it was taste—choosing constraints and saying no to clever features that add surface area but no leverage.

Clarity beats horsepower. If I can’t explain the job in one paragraph, I’m not ready to prompt anything.

Narrow beats general. Small, paid usefulness compounds faster than broad, free cleverness.

Edges create trust. Error states, permissions, handoffs. Decide them early.

Metrics must be boring. Minutes saved. Tickets deflected. Decisions sped up.

Thiel talks about “secrets” hiding in plain sight. Jobs insisted on ruthless constraints (“1,000 songs in your pocket”). Bezos banned slide decks because narrative exposes fuzzy thinking. I borrow all three: find the overlooked pain, box the problem tightly, and write the memo before I write the code.

We keep the Value Proposition Canvas (VPC) on one page: Customer Jobs, Pains, Gains, and Products/Services → Pain Relievers / Gain Creators. Then we translate that into a Quantifiable Value Proposition (QVP)—a single sentence we can measure:

For [segment], [tool] reduces/increases [metric] by [amount] within [timeframe] at [cost].

CareHub QVP (measured)

For property teams, CareHub reduces duplicate tickets and blind escalations, aiming for a lower median time‑to‑resolution and fewer off‑channel pings per week—tracked per building.

Awards QVP (measured)

For awards organizers, our voting system improves vote integrity and auditability—targeting high verification rates, low anomaly flags, and instant, exportable tallies.

Marketing Portal QVP (proven)

For growth teams, the portal lifts conversion by aligning creative and spend decisions—$2.6M in closed sales in 60 days on $30k ad spend in our own deployment.

The VPC keeps us honest about the job; the QVP forces a number we can defend. If a tool cannot earn a line on someone’s P&L, it is a toy.

Auth & roles: hosted auth or Postgres auth; supply a role matrix and let AI wire checks/tests.

Data layer: Postgres first; migrations from a schema doc; boring names win.

CRUD & tables: scaffold list/detail/edit with validation and empty states.

File storage: object storage + signed URLs.

Background jobs: queues for parsers, retries, webhooks, nightly rollups.

Integrations: start with one source of truth—CSV, API, or inbox—and normalize.

Retrieval for AI: keep models simple by feeding the right docs and examples.

Observability: log every automated decision; add audit trails day one.

Human‑in‑the‑loop: explicit escalation and a clean return path.

We are not reinventing frameworks; we are composing them with taste.

Day 1 — Memo & map. Two pages: user, pain, “done,” anti‑goals. One sequence diagram.

Days 2–3 — Walking path. Auth, database, and one end‑to‑end flow with fake data.

Day 4 — Real data. Parse the export or hit the API. Delete the mock.

Day 5 — Edges. Errors, retries, permissions, logs, and human handoff.

Day 6 — Docs, demo, deploy. Ship to first users; record three changes for next week.

We pivot often, but pivoting is not flailing when the compass is steady: remove pain today, keep the system small, and let compounding do its work.

“Isn’t this commoditized? Anyone can call the same model.” Yes—and irrelevant. The moat is intimacy with the workflow. Stand where the friction lives: the midnight CSV, the nine‑click vendor portal, the status meeting that should not exist. That is where wedges start.

“Won’t AI make everyone equally fast?” Power tools amplify differences. With taste and proximity to the problem, you lap teams with larger budgets who are still pitching platforms no one asked for.

I want to build a unicorn—not for the label, but because unicorns emerge when usefulness compounds long enough. The shape is predictable: a stack of small tools that matter, glued together by taste, hardened by real edges, and proven with numbers. AI accelerates assembly; accountability and standards stay with us.

If you are building now, try this this week: sit with one user, write the paragraph that explains their job in plain language, map the VPC, convert it into a QVP, name the parts that already exist, and let AI assemble the first walking path. Ship the smallest thing that makes tomorrow easier. Then do it again.

Build things that matter. The rest follows.