The hour before the call.

Every rep with a real book of accounts loses the same hour before every call. Company site, funding news, the buyer's profile, last quarter's press, a half-remembered competitor. Eight tabs, to learn enough about one account to write four sentences that do not sound like a template.

I lost that hour for two years, carrying a quota, and then watched ten people I managed lose it every day as well. It is not a skill problem and it does not go away when you get good. It is unpaid research work, sitting between a rep and the only part of the job anyone is actually measured on.

The signal that made me take it seriously

So I built tools for my own team: one that did the account research and drafted the first touch, one that tracked pipeline. Both were adopted organically beyond my own team, by people I had not pitched and did not manage.1Planning docCLAUDE.md §0Sales-first operator... promoted to lead a 10-person outbound team in 20 months. Built and shipped two internal AI tools (buyer research + first-touch drafting; pipeline tracking) that were adopted organically beyond his own team.

That is the part worth pausing on, and it is the closest thing to real demand evidence I have. Internal tools do not spread on their own. Reps are busy, cynical about process, and actively hostile to anything that adds a step. When something crosses a team boundary without anyone selling it, the pain underneath is real and the existing options are not covering it.

Internal tools do not spread on their own. When one crosses a team boundary unpitched, the pain underneath it is real.

What Oryxs is

One input, and you get back an account brief and a first draft, before you have opened a tab. Every claim in that brief is either cited to a real source or explicitly marked as something the system could not establish. It never fills a gap with a confident guess, which is the single behaviour that makes AI research unusable for a rep who has to say those words out loud to a buyer who knows their own company better than you do.

We don't compete on contact data. We compete on what happens after you have it.2Planning docdocs/USP.mdWe don't compete on contact data. We compete on what happens after you have it.

That is a strategic position, not just a slogan. The large players in this category spent a decade and serious capital acquiring contact data, and out-spending them on that is not a plan available to anyone starting today. The layer after the export, turning a row in a spreadsheet into something a human would actually send, is the part all of them are weak at. A crowded, well-funded category is not a risk to be argued away here. It is the proof that the budget already exists and buyers already pay for adjacent tools. The open question was never whether this market is real. It is who takes the layer nobody has taken yet.

What I got wrong twice before this

Two products came before Oryxs: an AI-agent cost observability tool, and a local-business AI agent platform. Both were finished. Both worked. On 2026-08-15 I stopped all engineering, design and founder hours on both, effective immediately, keeping the code on disk rather than deleting it.3Board entry · 2026-08-15board/DECISIONS.md, D-002Stop. No further engineering, design, or founder hours go into either product, effective immediately. Neither is deleted, the code stays on disk, costing Rs 0/month, retained as a free option.

The reason is in the decision log, written at the time and not softened since: both had been built to a high technical standard before a single human said they would pay, which is the exact failure this company's own constitution names as its primary anti-pattern, and it had by then happened twice.4Board entry · 2026-08-15board/DECISIONS.md, D-002 reasoningBoth were built to a high technical standard before a single human said they would pay. That is the exact failure mode CLAUDE.md names as its primary anti-pattern, and it has now happened twice.

They worked, and that was the problem. Working software nobody agreed to buy is the most expensive thing a small team can produce, because it feels like progress the entire time you are making it. What is different here is not that I am more confident. It is that this time I am building against a problem I carried a quota through, and against demand I already watched move between teams without being pushed.

The boundary I set before writing a line of code

I have a full-time job, and Oryxs is built on my own hours, my own laptop, my own accounts. That situation is where a lot of founders quietly create a problem for themselves later, so I wrote the boundary down before starting rather than deciding case by case, when it would be convenient to decide differently.

Nothing from my employer crosses: no customer lists, prospect lists, CRM exports, call recordings or pricing sheets. Not as a seed, not as a test dataset, not anonymised.5Board entry · 2026-08-15board/DECISIONS.md, D-004, The Wall (employer name redacted in the quote)No data. No customer lists, prospect lists, CRM exports, call recordings, pricing sheets, or anything learned from an [employer] system. Not as a seed, not as a test dataset, not anonymised. No employer code, no employer accounts, no selling to people I met through them, no work during their hours. What travels with me is what I know: how outbound works, how buyers decide, what a good cold email sounds like. That is mine, it is a well understood distinction, and no employer owns it.

I intend to leave one day as someone they would hire again and gladly refer. In Indian B2B I will meet these same people for the next twenty years.

Where this actually stands

Oryxs has zero paying customers today. That is the only number this company treats as its north star, ahead of users, signups or lines of code.6Planning docCLAUDE.md §1, The Prime DirectiveThe one number that matters right now: number of paying customers. Not users. Not signups. Not GitHub stars. Not lines of code. Paying customers. Current: 0 I am saying that in the same document where I am making the case, because you would find out anyway, and because a founder who hides the denominator is telling you something about how they will report every number after it.

The sharpest objection to all of this is already in our own decision log, from the day the company started: the real risk is that this system becomes productivity theatre that feels like progress while zero customers are contacted.7Board entry · 2026-08-15board/DECISIONS.md, D-001 dissent@skeptic notes the real risk is that the system becomes a productivity theatre that feels like progress while zero customers are contacted. That critique is live, unresolved, and the thing I watch hardest.

Every decision here is written down, dated, and never deleted, only superseded. The wrong arguments stay visible above their corrections. That is the product's own discipline turned back on the company: if we ask a buyer to trust software because it cites its sources, the company had better run the same way.

Which is why this post cites its own. Turn on Check the work at the top and every sentence backed by a real file in this repository lights up. The sentences that stay plain are opinion, and you should read them as exactly that.

Nothing here is a promise that this works. It is a record of what was believed, and why, while the outcome was still unknown. The number is zero. When it changes, this page will not be quietly edited to make the route look obvious in hindsight. It will just be a page with an older date on it.