The operator profile — taste, telemetry, and the strengths underneath.

Why audit yourself now.

I have always written better than I have introspected. Strategy work rewards external clarity — name the problem, frame the choice, drive the room. It does not reward turning the same lens on yourself. So you spend twelve years sharpening one half of the muscle and assuming the other half is fine.

104 days into running my own studio, the assumption stopped being safe. Two ventures, three brands, six distribution channels, one operator, and a stack that produces enough data about how I work that I can finally read myself the way I would read a target on a CDD. So that is what this essay is. Three sections. Taste. Telemetry. Strength.

---

① Taste. What I gravitate toward, before I think about it.

Taste is the part of you that ships before the deliberation does. It shows up in the tools you reach for, the colour palette you keep choosing, the sentence shape that sounds right in your head. It is the most legible thing about an operator and the hardest to fake. Here is mine, plainly.

A.Editorial over interface.

Every product I ship — Frollie Pro, this portfolio, the lecture site, the OSS detail pages — leans on the same vocabulary. Newsreader for the serif. Inter Tight for the sans. JetBrains Mono for numerals and roles. A deep phthalo green and a copper accent. Em‑dashes carrying the argument. Numbered tags in Roman or Arabic, never both. Pull quotes that double as load‑bearing claims. The aesthetic is not chosen — it is what I keep arriving at when I stop choosing.

The throughline: I believe a product is a piece of editorial. The dashboard is not a UI; it is a page. The repo README is not boilerplate; it is the lede. The CDD deck I wrote at McKinsey was a long‑form magazine article disguised as PowerPoint, and Frollie Pro is the same magazine article disguised as a Convex app.

B.Decomposition over abstraction.

I do not reach for clever frameworks. I reach for the issue tree. (See From issue trees to subagents for why those are now the same object.) Three‑part structures keep showing up because three is the smallest number that proves a pattern. Mechanic / driver / dad / builder. Spec / implementation / review / test. Domain / taste / AI fluency. The number of sub‑heads in any of my essays is almost always three or four. Fewer feels under‑specified; more feels like I am hiding behind a list.

C.Concrete numbers over rhetoric.

I would rather write "23 million tokens, USD $1,500, 90 days" than "transformative gains in development velocity." The strategy world taught me that any sentence without a number is a sentence the room can disagree with politely forever. Numbers settle arguments and they keep me honest. The stack panel on the homepage is the version of this discipline turned outward — every claim a number, every number sourced from the same repo I am writing from.

D.Static HTML over framework bloat.

This portfolio is hand‑written HTML with inline JSON manifests hydrated by twelve lines of vanilla script. No build step. No SSR. No framework. The posture is deliberate — when the editorial is the product, the page is the product, and the page should not be three abstractions away from the words. Frollie Pro gets the framework treatment because it has to. Writing does not.

E.One operator over coordinated team.

The deepest taste, the one I have only recently noticed: I would rather build the whole thing myself with agents than coordinate three people to build the same thing in twice the time. This is not pure preference — it is also a P&L decision (twelve‑person product team is AUD $2.5M/year; agent stack is USD $700/month). But the preference came first. The numbers ratified it later.

---

② Telemetry. What the numbers actually say.

The stack panel on the homepage is the cleanest dataset I have on myself. Five tools, four footer stats, and a year of build history sitting underneath. I will not re‑print the panel. I will read it.

Five readings, in order of how surprising I find them.

Reading 1The cadence is sustained, not bursty.

2,367 commits over 101 days is roughly 23 commits per day, every day, including weekends and holidays. That number is impossible without agentic delegation. A solo human at peak focus does six or seven commits a day on a good day, three on a normal one. The throughput here is the agent stack absorbing the boring half of the work and letting the human stay at the spec‑and‑review layer where the leverage actually is. The cadence is what an AI‑native operating model looks like in git.

Reading 2The skill count proves the topology.

62 custom skills in GSD is not decoration — it is a working sub‑agent topology. Discuss → plan → execute → verify, with twelve to fifteen specialised review and audit gates layered through. Each skill is a specific reps‑into‑muscle‑memory move I have wanted enough times to write down. The number is the closest thing to a senior‑engineer rubric I have for myself: every skill is a thing I now do by default that I used to forget to do.

Reading 3The surface is wide; the operator surface is thin.

337 modules in Convex. 8 systems in Frollie Pro. 32 features. The ratio matters: hundreds of internal modules collapsing into eight things the operator actually touches. Wide surface, thin operator surface — that is the inverse of how most B2B SaaS gets built (thin internal architecture, wide operator surface, escalating support load forever). The architecture taste is paying off in operator simplicity. We answer fewer support tickets in a quarter than HelloFresh ANZ did in a day.

Reading 4The token spend is a rounding error on the marketing budget.

36.8M tokens at the Sonnet 4.6 / 4.7 mix is roughly USD $700–$800 per month. Frollie's monthly marketing spend on a single SKU launch is multiples of that. The agent stack does not appear in the P&L as a cost line — it appears as a category change. (I made the same point at length in 60 days, 23 million tokens; the new data point is that the next 44 days held the ratio.)

Reading 5The repo is the company.

Five apps on one Vercel team. One Convex deployment. One GitHub org. The operating system, the consumer site, the lecture site, the portfolio, the OSS — all in one weave. There is no "engineering org" separate from "the business." The repo is the business. Source control is the org chart. The PR queue is the priority list. That is not a slogan; it is a posture, and the data says I am living it.

---

③ Strengths. What taste plus telemetry compose into.

Strengths are the place taste and telemetry meet. Taste says what you would do; telemetry says what you have done; the overlap is where you are actually strong. Five fall out of the audit, in order of how much advantage I think they confer.

Strength 1Three‑career composability.

Seven years at McKinsey teaching me what good engines look like. Two years at HelloFresh ANZ teaching me what it costs to drive one. Eighteen months as a builder finding out whether I actually know what I am doing. Most operators have one of these toolkits. Some have two. The composition of all three — diagnostic, P&L, agentic build — is the rarest of my advantages, and the one that lets me compress what would be a six‑person leadership team into a single seat. (I wrote this up at length in Mechanic, driver, dad, builder; this is the audit version of the same claim.)

Strength 2Senior judgement at agent speed.

Taste is calibration to a specific audience, and twelve years of reps calibrate the reviewer in me to spot wrong‑shape work fast. The agent stack produces confidently wrong answers at speed. I produce confidently right reviews at speed. The pairing is the new senior contribution. (See Taste vs. execution engine for the underlying argument.) Where I outperform a junior plus an agent is not in writing the code — it is in not approving the code that should not ship. The 65% of token spend that goes to review and refactor is the part the seniority actually buys.

Strength 3Decomposition discipline.

Issue trees scale to sub‑agents. The brain that broke down a McKinsey problem into a five‑week workplan now breaks down a Frollie feature into a spec → implementation → review → test topology. The same instinct, ported. The 62 custom skills are the externalised version of this — every skill a discrete decomposition step I have wanted to repeat. The discipline is unromantic and the most leveraged single thing I do.

Strength 4Editorial as part of the work.

I write while I build. The writing is the positioning, the recruiting tool, the diligence brief, and the public artefact that compounds while I sleep. Eleven essays in twelve months on this site, each one a load‑bearing piece of the brief I send investors and operators. The output is not a side project — it is a moat. Most operators write nothing or hire a content function and produce ghosted thinness; the writing on this site is mine, dated, and worth referencing two years later.

Strength 5Bootstrapped patience.

90 days from zero to profit at Frollie. No external capital. No engineering hires. No board to placate. The patience is the part most founders cannot mimic because the muscle was built somewhere else — for me, McKinsey weeks where the answer would not arrive on Tuesday and I had to stay calm about it through Thursday. Patience is a strategy under‑rated and a temperament under‑described. The data says I have it.

---

What the audit does not say.

Three things I am genuinely watching.

First, the throughput is sustained so far. 101 days is not a year. The cost of the agentic stack is bounded by my own attention, not by the API spend. I will know whether this cadence holds at the 365‑day mark, not the 104‑day mark.

Second, the moat is my taste, not the taste. If I get hit by a bus, the stack keeps running for about thirty days before the review function decays. That is a real concentration risk and one I think about when I write. Documenting taste is not the same as transferring it.

Third, the strengths above are mostly internal. Distribution — getting Frollie in front of the right buyers, getting IKIGAI in front of the right operators — is the part of my next 365 days that the audit cannot grade yet. Build capacity is no longer the bottleneck. Direction is. (See Force × direction.) The next audit will be about whether I pointed the cadence in the right direction.

Where this leaves me.

Taste, telemetry, strength. What I gravitate toward, what the data says, what the overlap composes into. Writing it down is the closest I have come to reading my own diligence pack. The advantages are the three‑career arc, the review function, the decomposition reflex, the writing flywheel, and the temperamental patience. The risks are sustainability, transferability, and direction.

I sit, again, at 2am while Ava sleeps. The agent stack is committing while I write. The next audit drops in another 100 days. The advantage I most want to compound between now and then is the one the data cannot yet see — the brief I have not yet taken, the venture I have not yet started, the next first sentence of the next first paragraph.