Writing
What we learned building this.
Notes on validating an idea, paying for AI by the token, and answering the phone without hiring. Specific where we can be, and honest about what is not built yet.
- Published Aug 22, 20264 minutes to read
A Typeform alternative for developers
A smaller one-question-at-a-time form tool, what it costs against Typeform's published prices, and the features it does not have.
- Published Aug 22, 20265 minutes to read
A waitlist signup is not demand
What an email address in a box actually tells you, what a stated price does not, and how to collect the evidence that survives contact with a launch.
- Published Aug 22, 20267 minutes to read
What an AI call center actually costs per minute
The arithmetic behind a $0.23-a-minute AI phone line, the charges that sit outside it, and how to work out whether it beats what you pay now.
- Published Aug 22, 20265 minutes to read
An agentic IDE is not a chat window
What changes when the model can act — approval surfaces, undo, cost reservation — and the four things Codoro deliberately will not let an agent do.
- Published Aug 22, 20265 minutes to read
Answering service or AI phone agent
How the two are priced, which call patterns favour each, and the four questions that decide it before you look at a single rate.
- Published Aug 22, 20265 minutes to read
Asking about price on a signup form
How to word the question, why brackets beat a free-text box, and the three readings of the answers that are worth anything.
- Published Aug 22, 20264 minutes to read
Answering the phone in Spanish without hiring for it
What language detection actually does on a call, the five language hints the product accepts, and where a second language stops being a software problem.
- Published Aug 22, 20265 minutes to read
Your own API key, or the vendor's credits
What a 40% markup pays for, what zero markup costs you in effort, and the security difference between a key in a keychain and a key on a server.
- Published Aug 22, 20265 minutes to read
Disclosure when an AI answers your phone
What a recording notice is for, why storing nothing does not settle the question, and the parts of this you must ask a lawyer rather than a vendor.
- Published Aug 22, 20265 minutes to read
Capping what an AI agent can spend on one task
Why a daily budget arrives too late, how a reservation has to be taken before the price is known, and what a cap should do when it binds.
- Published Aug 22, 20265 minutes to read
Conditional logic without a builder you cannot read
Five operators instead of twenty, the three condition mistakes that look correct and are not, and why forward references are the dangerous one.
- Published Aug 22, 20265 minutes to read
Connecting a Twilio or Telnyx number to an AI agent
Why bring-your-own-SIP beats a bundled number, what the carrier still owns, and the failure modes worth knowing before you point a live line at anything.
- Published Aug 22, 20265 minutes to read
Customer research forms people actually finish
The six questions worth asking, why every one of them is about the past, and the ordering that stops people leaving at question two.
- Published Aug 22, 20264 minutes to read
Double opt-in and the signups that are not real
The four kinds of fake signup a public form attracts, why verification is a data problem before it is a deliverability one, and what it costs you.
- Published Aug 22, 20264 minutes to read
Embedding a waitlist in a site you already have
Hosted page, embed, or your own form against the API — what each costs you in control, and the export trap that silently loses rows.
- Published Aug 22, 20265 minutes to read
Four products, one account, and why
What four separate sites on one login actually buys, where the shared spine helped, and the two products that are not purchasable yet.
- Published Aug 22, 20264 minutes to read
How many waitlist signups is enough
Why the number you are looking for does not exist, the four figures that replace it, and how to set a bar before you can rationalise past it.
- Published Aug 22, 20265 minutes to read
How to review a diff you did not write
Why AI-generated changes fail differently from human ones, the order to read them in, and the three failures that survive a careful read.
- Published Aug 22, 20266 minutes to read
One question per screen, and why people finish
The mechanism behind single-question forms, what a half-finished response tells you, and an honest note about the completion data we do not have.
- Published Aug 22, 20264 minutes to read
Referral mechanics that actually move a queue
Why position-jumping works, what it does to your data, and the abuse you will get within a day of turning it on.
- Published Aug 22, 20264 minutes to read
Switching models in the middle of a task
Why one model for a whole task is usually the wrong default, what a switch actually costs, and the case for never doing it silently.
- Published Aug 22, 20265 minutes to read
The gap between a landing page and a signup
Where visitors actually stop, why the fix is nearly always the words rather than the button, and how to tell which half is broken.
- Published Aug 22, 20265 minutes to read
The honest math of our affiliate program
Every rate and cap in the table, why one product pays a tenth of another, and the case for not bothering with the low ones.
- Published Aug 22, 20264 minutes to read
Tokens, explained for people who just want it to work
What you are actually paying for when an AI writes code, why the bill grows faster than the work, and the four numbers on a rate card.
- Published Aug 22, 20265 minutes to read
Validating an idea without writing code
What you can find out before building, what you cannot, and a three-week sequence that produces facts rather than encouragement.
- Published Aug 22, 20265 minutes to read
Validating form answers in the database, not the browser
Why client-side rules are a UX affordance rather than a guarantee, what a constraint buys that a check does not, and the ceilings that stopped request 181.
- Published Aug 22, 20265 minutes to read
Vibe coding without losing the plot
How a working app becomes one you cannot change, the four habits that prevent it, and where the money goes when a session starts going wrong.
- Published Aug 22, 20265 minutes to read
What a half-finished form tells you
Why storing partial answers beats measuring a completion rate, the three reasons people stop, and reading a drop-off cluster correctly.
- Published Aug 22, 20266 minutes to read
What an AI phone agent cannot do
The limits worth knowing before you put one on your main line — call caps, no outbound, no transcripts, and the integrations that do not exist.
- Published Aug 22, 20265 minutes to read
What a missed call actually costs
How to work out the real number for your business from your own phone bill, and why the published statistics on this are not worth quoting.
- Published Aug 22, 20265 minutes to read
What shipping in three languages actually cost us
153 KB of gzipped catalogue on every route, a tree-shaking assumption that was wrong, and the four i18n failures that pass every test.
- Published Aug 22, 20264 minutes to read
What to email a waitlist, and when
A sequence that keeps a list warm without inventing news, why replies are the only metric worth reading, and what is not built yet.
- Published Aug 22, 20265 minutes to read
When an AI phone agent should give up and transfer
The four signals worth transferring on, why our agent cannot choose who it transfers to, and what a five-minute call cap does to the decision.
- Published Aug 22, 20266 minutes to read
Why your AI coding bill is so high
Where the money actually goes in an agentic coding session, why estimates run under, and the difference between reserving a cost and settling one.