Building “Solomon”: An Internal Agent, and What V1 Taught Us About UX

“Custom AI assistant” spiked to 1,002 U.S. searches in January 2026 — nearly 3x the trailing months — on a real SERP almost entirely built from individuals documenting their own weekend build. Here is what shipping a production internal agent, not a hobby project, actually taught us.

July 13, 2026 · 6 min read · David George
What we solve

How much of your team’s institutional knowledge lives in one person’s head?

90

conversions a month you’re likely flying blind on — and optimizing against.

The emergence The commercial pull Who’s competing for attention Growth or decline How PPC Snobs executes here The emergence The commercial pull Who’s competing for attention Growth or decline How PPC Snobs executes here
Quick answer

A custom AI assistant is a purpose-built agent wired into a specific team’s own tools and data, rather than a general-purpose chat interface. We built ours — internally, we call it Solomon — to hold institutional knowledge that otherwise lives in one person’s head, and shipped it as production infrastructure, not a demo.

TL;DR
  • Real U.S. demand for “custom ai assistant” spiked to 1,002 searches in January 2026 — nearly 3x the trailing months — before settling back to 338 by July.
  • The real top five is almost entirely DIY: a Substack post and two Reddit threads and a Medium build-log, average Domain Rating 89 — genuine authority, but from individuals, not vendors.
  • Only one real product shows up in position four: CustomGPT.ai at DR 74 — the sole company actually selling a “custom AI assistant.”
  • KD reads a moderate 26, but the real fight is not a product battle — it is against thousands of people publishing their own weekend build.
  • Solomon is not a weekend build. It is a production internal agent, and V1 taught us a UX lesson we are willing to publish.

Type “custom AI assistant” into Google and you will mostly find people who built one over a weekend, writing it up. That is a real, useful category of content — and it is not what we built when we shipped Solomon.

The emergence

Real demand is thin but genuinely spiky: 302 U.S. searches in July 2025, a January 2026 peak of 1,002 — nearly three times the trailing months — before falling back and settling at 338 by July 2026. That spike looks like a holiday-season wave of people deciding this was their New Year build project.

338
current US searches / mo
1,002
January 2026 peak
600
global searches / mo
Source: Ahrefs, US, Jul 2026

The commercial pull

A modest $0.90 CPC fits a DIY-heavy audience — people comparing free tools and open-source approaches, not evaluating a five-figure vendor contract. The commercial opportunity here is not the click; it is the credibility of having actually shipped one.

Who’s competing for attention

The real top five is almost entirely individuals: a Substack build-log at DR 94, two Reddit threads at DR 95, a Medium write-up at DR 94 — and exactly one actual product, CustomGPT.ai, holding real position at DR 74. This is a page owned by people, not companies.

Who owns real organic position for “custom ai assistant” (Domain Rating)
Reddit build threads95
Solo-builder Substack/Medium94
CustomGPT.ai (the one product)74
Source: Ahrefs SERP overview, US, Jul 2026

Growth or decline

Structurally small with one sharp, unexplained spike. We are reporting the January 2026 anomaly as measured, not modeled — whatever drove it did not repeat, and the trailing trend since has been a steady, quiet 250–400 a month.

A weekend build vs. Solomon
Weekend buildSolomon (production)
Where it livesA laptop, a demo linkDeployed infrastructure, on-call
What it knowsWhatever you fed it onceLive-synced to real systems
Who trusts itThe person who built itThe team, after V1’s UX fixes
What breaks itNothing — nobody depends on itEverything, because people do

How PPC Snobs executes here

We built Solomon to answer the questions that otherwise interrupt one specific person all day — and V1 taught us that answering correctly was not enough if nobody trusted the answer yet. We shipped a second version focused entirely on showing its work, not just its conclusion.

“The hard part of an internal agent was never getting it to be right. It was getting the team to believe it, before they double-checked with a person anyway.”
DG
Article by

David George

David leads the build side of PPC Snobs, shipping custom Claude MCP connectors on Firebase and Cloud Run — including the QuickBooks integration that reconciles ad spend to revenue in the client’s own ledger.

FAQ

Questions, answered.

An AI agent built and wired specifically into one team’s own tools, data, and workflows — as opposed to a general-purpose assistant with no access to your systems.

From the author

Why this matters.

David George on the thinking behind it.

DG
David George
Chief Technology Officer

Most growth problems aren’t a channel problem — they’re a seam problem. The money leaks between measurement, pages, and media.

DG
David George
Chief Technology Officer · PPC Snobs

I won’t sell you three vendors who blame each other. One team, one source of truth, one number that’s actually real.

DG
David George
Chief Technology Officer · PPC Snobs

Buy the engine, not the ads. The ads are the easy part — the system underneath is where the compounding lives.

DG
David George
Chief Technology Officer · PPC Snobs
Pricing

Investment scales with ambition.

Two ways to engage. Both transparent — no SDR follow-ups, no proposal theatre.

Self-serve

Build your own retainer

Pick the modules you need. See exact one-time and monthly investment before you commit to anything.

Live total calculator
Modular pricing — no bundles
AI-enable, then scale on agents
Open the configurator →
RecommendedWhite-glove

Request a custom quote

For complex stacks, multi-brand portfolios, or projects above $50K/mo. Scoped on a call, priced on a doc.

Architecture audit included
Quarterly business review
Dedicated account manager