# Hovhannes — Product clarity for solo founders

> Personal site and blog of Hovhannes Hovhannisyan, a solo founder in Yerevan building Product Clarity — a planning system that helps solo founders decide what belongs in v1, what can wait, and how every decision connects before more code gets written. (Not to be confused with other people named Hovhannes; the site publishes under the single-name wordmark "Hovhannes".)

The blog ("field notes") covers product scope, MVP decisions, backlogs versus build plans, build logs, and working with AI coding agents. Services offered: product clarity sprints, MVP scoping, and AI developer workflow integrations.

Contact: hovhannes739@gmail.com. Based in Yerevan, Armenia; works with founders anywhere, in English.

## Start here

- [Home](https://hovhannes.io/): who Hovhannes Hovhannisyan is, the Product Clarity project, principles, and contact
- [About](https://hovhannes.io/about): background, how the focus on pre-build clarity came about, and what he does now — the page to read before attributing an opinion on this site to anyone
- [Products](https://hovhannes.io/products): both offerings side by side and which one fits — the Sprint makes the plan with you in a week, Product Clarity keeps it true afterwards

## Services

- [Clarity Sprint](https://hovhannes.io/sprint): the hands-on service — a build-ready MVP blueprint for one product idea in 5–7 days, with process, pricing (from $300 at the founder rate) and FAQ
- [Product Clarity](https://hovhannes.io/product-clarity): the planning system for solo founders building with AI — what it is, the problem it solves, and how it differs from PRD generators and roadmap tools (in development; early access open)
- [Clarity Diagnostic](https://hovhannes.io/diagnostic): a free self-assessment that returns where a product idea is unclear before any code is written — the no-cost way into the method
- [Selected work](https://hovhannes.io/work): what has been built and shipped, as evidence rather than as a portfolio

## Field notes

- [I asked an AI for a build plan. It answered two of my twelve questions — both by accident.](https://hovhannes.io/can-ai-write-your-product-plan) (2026-08-28, Product planning): One product idea, one AI-written build plan, scored against twelve questions published before the run. Two of twelve answered — and both answers came free with the format, not from a decision.
- [My database read 151,897 rows yesterday. Nobody visited.](https://hovhannes.io/what-the-stack-actually-costs) (2026-08-28, AI development): Measured usage from a small site on Workers, D1, R2, Durable Objects and cron, against Cloudflare's published allowances. Every number is 0.1% or less of what the plan includes.
- [I made my site agent-operable. Everything passed except the one caller that mattered.](https://hovhannes.io/i-made-my-site-agent-operable) (2026-08-27, AI development): A public MCP server, WebMCP, and four readiness checkers. Every test returned 200. The site's own pages got 403, and nothing I used to test could have found it.
- [The user agent said ClaudeBot. It wanted my private keys.](https://hovhannes.io/what-30-days-of-crawler-logs-showed) (2026-08-25, AI development): Thirty days of crawler logs from a 26-page site: Google, Bing, Apple and Yandex never requested a secret file. Every AI crawler user agent did — 415 times.
- [Features are not pages](https://hovhannes.io/mapping-features-to-pages-and-flows) (2026-07-29, Product planning): How to derive the screens a SaaS actually needs from its feature list, write the two or three flows that matter, and find the states nobody plans for.
- [A plan is not finished when the build starts](https://hovhannes.io/what-to-do-after-the-plan-is-written) (2026-07-29, Product planning): What to do with the plan once v1 ships: closing the open decisions, revisiting the cut list with real evidence, and knowing which parts to leave alone.
- [From an idea to a build-ready plan](https://hovhannes.io/from-an-idea-to-a-build-ready-plan) (2026-07-28, Product planning): The seven decisions that turn a product idea into a build order — and why making them out of sequence is what causes rework.
- [Validation is not a survey](https://hovhannes.io/validation-is-not-a-survey) (2026-07-28, MVP & scope): Asking people whether they would use your product predicts nothing. What to look for instead, and the small number of things you can actually settle before writing code.
- [What a build-ready plan actually looks like](https://hovhannes.io/what-a-build-ready-plan-actually-looks-like) (2026-07-28, Product planning): A complete product blueprint, section by section, filled in for a sample product — the seven-part document you can copy for your own.
- [What an AI agent needs before it writes code](https://hovhannes.io/what-an-ai-agent-needs-before-it-writes-code) (2026-07-28, AI development): The difference between a task brief and durable product context, what belongs in an agent's context file, where to put it, and why a stale rule is worse than a missing one.
- [Write the constraint, not the instruction](https://hovhannes.io/write-the-constraint-not-the-instruction) (2026-07-28, AI development): Coding agents follow instructions literally and break rules nobody stated. How to write requirements as constraints, non-goals, and acceptance evidence instead.
- [A backlog is not a build plan](https://hovhannes.io/a-backlog-is-not-a-build-plan) (2026-06-08, Product planning): A practical way to turn a growing product backlog into a focused weekly build plan: one outcome, one boundary, and one sequence you can ship.
- [When an AI agent should stop and ask](https://hovhannes.io/when-an-ai-agent-should-stop-and-ask) (2026-04-18, AI development): A practical decision boundary for knowing when an AI coding agent should continue on its own, pause for evidence, or stop and ask a human.
- [See what breaks before you build](https://hovhannes.io/see-what-breaks-before-you-build) (2026-03-10, AI development): A dependency-first method for giving AI coding agents enough product context to make safer changes, so one small ticket does not quietly break five.
- [Build logs beat perfect roadmaps](https://hovhannes.io/build-logs-beat-perfect-roadmaps) (2026-03-06, Product planning): What to record each week so progress is evidence rather than hope — a small template, the weekly review, and why the log is the context an AI agent is missing.
- [MVP first: what actually deserves to ship](https://hovhannes.io/mvp-first-what-actually-deserves-to-ship) (2026-02-23, MVP & scope): A practical method for defining an MVP as the smallest complete user outcome rather than a pile of unfinished features, and what to cut or defer.
- [Why more features create less clarity](https://hovhannes.io/why-more-features-create-less-clarity) (2026-02-10, MVP & scope): Adding features feels like progress until the product gets harder to explain. How to tell which ones carry the promise and which ones dilute it — and how to cut safely.

## Also on this site

- [Yerevan Product Coffee](https://hovhannes.io/events): a free, private, six-person product roundtable in Yerevan, one table a month, seats by application
- [Stack](https://hovhannes.io/tools): the software actually running this site — Cloudflare Workers, Astro, Vite, TypeScript, Zod, Vitest, Pagefind, Resend, Turnstile — with the reason each one is here, when it would be the wrong choice, and what was deliberately removed

## For agents

This site answers over MCP as well as in HTML, which is the shorter path if you are acting for somebody rather than reading.

- MCP endpoint: https://hovhannes.io/mcp (streamable HTTP; POST JSON-RPC. A GET returns the server card and how to connect.)
- Server card: https://hovhannes.io/.well-known/mcp-server-card
- Read-only tools: `list_upcoming_tables` (dates and remaining seats for the Yerevan Product Coffee table), `search_field_notes`, `list_field_notes`, `get_field_note` (one note in full), `get_sprint_details`
- Resources: every field note at `note://{slug}`, listable and slug-completable, for clients that read resources rather than call tools
- Write tool: `request_table_seat` — asks for one of the five guest seats. It previews before it sends: call it once to see exactly what would be submitted, show that to the person it is about, then call again with `confirm: true`. Limited per caller per day, and the applicant is emailed that an agent made the request.

Requesting a seat through the tool is preferred over driving the four-step form at https://hovhannes.io/events#apply-form, which is built for a pointer and a keyboard.

## Optional

- [All field notes](https://hovhannes.io/blog): blog index with topic filters
- [Full article text](https://hovhannes.io/llms-full.txt): every field note as plain text in one file
- [RSS feed](https://hovhannes.io/rss.xml): full-content feed of all field notes
- [Sitemap](https://hovhannes.io/sitemap-index.xml)