A Team's Quarter in One Evening a Day: Building TeleTone with Claude Code
Nov 9, 2026 · 7 min
Every project I've estimated in 20 years started with the same arithmetic: scope, people, months. When I sketched TeleTone, the numbers came out where they usually do for a product like this: 3–4 months for a team of 5–6 — a backend engineer or two, a frontend developer, a data/ML person, DevOps, and someone holding product and QA together.
I built the working prototype alone, in one month, in 2–3 hours a day — evenings, after everything else in my life had taken its share.
This post is not about the architecture. It's about what changed in how one experienced engineer can work, and why I think it matters for anyone deciding whether to hire a team or validate an idea first.
What TeleTone is
TeleTone is a Telegram channel checker for ad buyers. It answers one question: is this channel worth paying for an ad post? Paste a channel, and you see real reach, engagement, signals of fake views or bought subscribers, and who else has advertised there.
Telegram advertising is a large market with a trust problem. Subscriber counts are cheap to inflate, views can be bought, and buyers often pay based on a screenshot and a promise. TeleTone exists to replace the promise with data.
What actually got built in a month
Not a landing page with a Figma mockup behind it. A running system:
- Data collection from public Telegram channels: history backfill, daily snapshots, tracking of edits and deletions
- A web app with channel reports, an authenticity check, user accounts, plans and quotas, in English and Ukrainian
- An admin panel for customers, workers and content analysis
- AI analysis — topics, summaries, sentiment, toxicity, channel digests, Q&A over a channel — running on local open models, not paid APIs
- An MCP server, so AI agents can query the archive directly
- A separate marketing site with an early-access flow
- CI/CD building and deploying everything automatically
- A competitive analysis of the main players, turned into an 82-row feature matrix with an MVP scope, acceptance criteria and pricing hypothesis
It runs on a small VPS plus hardware I already own at home. The monthly infrastructure bill is about the price of a dinner.
The numbers, honestly
A team of 5–6 for 3–4 months is roughly 15–24 person-months, or somewhere around 2,500–3,800 working hours.
My month was 60–90 hours.
That's not a 2× or 3× improvement. It's an order of magnitude, and more. I want to be careful with that claim, so here is what it does and doesn't mean:
- It's a prototype heading into a paid beta, not a product that has survived thousands of users. Scaling, support and hardening still take time.
- The estimate is mine, made the way I'd make it for a client — before I knew how it would go.
- The speed didn't come from cutting corners. That part matters most, so it gets its own section.
Best practices didn't get skipped — they got cheap
The usual way solo projects go fast is by dropping discipline: no docs, no CI, secrets in the code, "we'll fix it later." I've cleaned up enough of those projects as a consultant to know where that ends.
The surprise with Claude Code was the opposite: the things that normally slow a small team down became almost free, so I did all of them.
- Scope before code. Competitor research, a feature matrix, a written MVP definition with a user story and acceptance criteria. On a normal project that's a week of a PM's time. Here it was an evening.
- Decisions written down. Architecture notes, risks with mitigations, open questions — kept current, not written once and abandoned. Documentation stops being a chore when updating it costs a sentence of instruction.
- Automation from day one. Every change goes through a pipeline that builds and deploys. No manual "copy files to the server."
- Security by default. Least-privilege database roles, nothing at home exposed to the internet, secrets out of the repo, public channels only, read-only access to Telegram.
- Observability. Every background worker reports its health; if something goes quiet, I get an email. When you only have two hours a day, you can't afford to spend one of them figuring out what broke.
- Small, reviewable steps. Each session had one clear goal, ended with a working state, and every change was reviewed before it went in.
None of this is new. What's new is that the cost of doing it right fell to roughly the cost of doing it wrong.
Where the leverage really comes from
People assume AI makes you faster because it types code faster. That's the smallest part.
The bigger win is that coordination cost disappeared. A team of six spends a large share of its time on standups, handoffs, waiting for reviews, re-explaining context and resolving misunderstandings between frontend and backend. With Claude Code, the "team" shares my context completely, works on whatever layer is needed in the moment, and never waits for anyone.
The second win is that research and rework got cheap. Trying an approach, discovering it's wrong and changing direction used to cost days. Now it costs an evening, so I'm willing to do it — and the result is a better design.
What Claude Code didn't do
It didn't decide what to build. It didn't pick the niche, choose to run AI on local models to keep costs near zero, or decide that the first version should answer one question well instead of ten questions badly. It didn't know which shortcuts would hurt in six months.
That's where 20 years of building fintech systems, running infrastructure and leading engineering teams paid off. My role changed from writing most of the code to being the architect, product owner and reviewer of a very fast, very patient engineering team.
AI doesn't replace experience. It multiplies it. A vague request gets you vague software. A precise one, from someone who knows what good looks like, gets you a system.
What this means if you're building something
If you're a founder with an idea, the old choice was: raise money and hire a team, or don't find out whether the idea works.
There's now a third option. One experienced technical lead with the right AI tooling can take an idea to a working, properly engineered prototype in weeks — something real users can touch, that you can show to investors, and that a future team can build on instead of rewriting.
That's the kind of work I do as a fractional CTO. If you have an idea you want to validate before committing to a full team, let's talk.
And if you buy Telegram ads and are tired of paying for subscribers who don't exist, TeleTone early access is open.