Engineers in a product company, not an agency and not an outsourcing shop. The product has already been in development for months or years, so the knowledge stays relevant.
Said in the session Outsourcing builds something and then maintains it or hands it over. That knowledge is more fragmented, and the early phase needs a stable product. People may change; the product should not.
Hypothesis to test The first design partners are smaller startups, who still need this. If that holds, the next partner can be a larger startup. After a second design partner, look at the later ICP: companies with 50–100 engineers.
Said in the session The bar he gave for that first partner, before the board started: at least 4–5 engineers, and at least six to twelve months of building. He also said naming the companies takes another day or two of asking people how the team is structured.
Hypothesis to test What would drop this segment: they say the team is small enough to know who does what, the product is small enough to fit in one agent’s context, and they already know the work.
There is no customer proof for this segment yet. Hackathon compliments were treated in the session as encouragement, not as a yes.
02 · Buyer and user
Programmers use it. A small startup’s CTO can also buy it.
The user is a programmer. In a smaller startup the buyer is the CTO, and that person may also be a user. In a larger company the buyer can be an engineering manager or a head of engineering.
Said in the session At SMG, his employer, contracts under 10,000 a month do not go through a procurement team. That was an example of how a purchase can happen. SMG was not named as a design partner.
Hypothesis to test Who to contact first is undecided. One path: the programmer feels it, becomes the champion, and pushes it upward. The other path: start with the buyer. In a startup those can be the same person, which is why the smaller company comes first. Bottom-up is the bet only where the user can actually move the buyer.
Smaller startup, first
Later, larger team
User
Programmer
Programmer
Buyer
CTO, sometimes also a user
Engineering manager or head of engineering
First message
Whichever of those he already knows
Not this round
03 · Problem
The agent’s performance, and what that performance costs.
The company-level pain is the performance and the cost of their agents. Under that: the models are not trained on the company’s data, and an agent with Slack access still cannot pull the fact it needs. The task then goes out without the company’s knowledge.
Said in the session He first described the mechanism. Asked for the pain at company level, he said agent performance, then performance and cost. He could not say how often it happens, or how urgent it is, without usage. A later guess he offered: if cost fell by half, or performance doubled, it would be urgent. Those numbers are not measurements.
Hypothesis to test Getting it right on the first try, without another prompt and without “go look here,” is a gain. He said that is not the main problem.
How often it hurts, and whether anyone will pay to stop it, is the point of the design-partner calls.
04 · Value proposition
Not locked. The session stopped on purpose.
“Improved performance” is not the sentence. A technical listener can hear it. An investor, or a CEO who is not in the infrastructure, cannot.
Said in the session He would rather leave it blank than invent a number. A quantified claim, such as hours saved, can only be a guess until a design partner produces data. The value proposition will change as the customer profile changes. It does not get nailed once.
Shaped from the session, not a new claim A sentence that stays inside what he said, still waiting for his edit:
Nucipal makes a product team’s coding agents perform better and cost less, because they can use how that company actually works, without someone maintaining a wiki.
The benefit he is aiming at is performance and cost. Time saved and fewer corrections stay the secondary gain until a call says otherwise. No percentage goes on this sentence.
05 · Positioning
Infrastructure for agents. Not another app, and not Glean’s contract.
The alternatives he named are Glean, chatbot-only agent products, and Notion-style docs. The difference he claimed is who it is for, what the buyer pays, and that the knowledge maintains itself and can be used by agents outside one chat app.
Alternative
What he said
Status
Glean
The largest one he sees. Started as enterprise search, now pointed at agents, with a large base and a head start on retrieval. He put them near 300 million ARR and an average contract around 60,000 euro. Hard to “be better” head-on. The opening is that they sell to a larger buyer than the first Nucipal segment.
His view in the session. The ARR and contract figures were not re-checked for this page.
Chatbot agent tools
Built for a defined assistant, such as support. Extra context when the answer is missing. Not built for the case where the agent is doing the work, with other people, rather than sitting beside it.
His distinction. No vendor named.
Notion, docs, slides
A person has to write and maintain them. Nucipal should update itself when something new is agreed, including in Slack. It should work with Codex, Copilot, or whatever the team already uses.
His distinction. “Updates itself from Slack” is a product claim still ahead of a demo.
Do nothing, or the current stack
Left for the calls. The session’s rule: competition includes the way they work today, not only a company with the same product.
To learn from Luka, not to assert.
Said in the session His matrix: manual versus automatic on one axis, application versus infrastructure on the other. Notion and ChatGPT company-knowledge are applications, with a human as the user. Nucipal sits on the infrastructure side, so agents in other tools can call it.
Hypothesis to test The moat, what would be hard for a better-funded team to copy, was asked and not answered.
06 · Channel
Founder-led. One channel at a time.
He starts the conversations himself. Whether the message goes by email or LinkedIn is the experiment. Both at once would make a weak reply impossible to read.
Said in the session He agreed to change one thing at a time: the segment, or the offer, or the channel, not all three. Whose network comes first — his, Milan’s, a hackathon, a conference — he said he would think through. That choice is not made.
Said in the session The same day he spoke with Daniela at Digitalna inicijativa Srbije, after meeting Tanja Kuzman. Nucipal is not raising from that angel group now. The ask was whether someone in their portfolio, or in Unlockit, could become an early design partner, or whether an angel could later mentor the fundraising. Nothing from that conversation is a partner yet.
Said in the session He wants more conversations than a short list, toward 15 or 20 once there is a concrete demo, so he can turn some away. Until that demo, the focus stays on the technical work, which is why the number stays lower for now. Deadlines have been missed because of other obligations. The session’s correction: keep numeric goals anyway, and record why a number was missed.
07 · Conversion
Qualify in the messages. Sell on the call. No demo yet.
The first exchange checks whether the problem is real. A pilot conversation comes after that. The product is not at a level where a demo is the opening.
Said in the session What the call is for: does this problem exist here, how often, how they handle it now, what is hard or expensive, which data they would share, and what smaller test is possible if they will not start a pilot. Success has to be defined before the number arrives, so “5% feels fine” cannot be decided afterward.
Said in the session His first draft to Luka congratulated him on Stitch, then said he is CEO of Nucipal, described company-knowledge infrastructure, and claimed a percent improvement in performance and a percent drop in cost. He did not have the ask. The correction in the session: the personal line stays; the product and the percentages wait. A few messages establish the problem. Only then say what Nucipal is, ask how they handle it, and invite 20–30 minutes. The sale is the call, not the text.
What “ready for a pilot” means as a minimum product was asked and not answered. Price after a pilot — subscription, annual, usage — was asked and not answered. For a design partner, “willing to hear more” is the signal he named. It is not proof they will pay.
08 · Evidence loop
One variable, then a written result.
Each conversation updates one step on this board. The segment changes if a team says they are small, they already know who does what, and the product fits in a single agent context.
Said in the session If twenty messages produce one reply, the open question is which of the three moved: the segment, the wording, or the channel. Changing all three hides the cause. The same discipline he described for the product — log the run, compare it with the previous run — is the discipline for these calls.
Nothing on this page is confirmed by a customer. A second version replaces a step only when a call supplies the replacement in their words.
Customer profile
Jobs, pains, and gains, as far as the session got.
This is the profile attached to the board. It is a hypothesis. The calls are supposed to replace these lines with their words.
Jobs
Get an agent to finish a task that depends on how this company builds.
Find the fact that currently lives in Slack or in someone’s head.
Keep that knowledge current without writing it into a doc.
Pains
Agent performance.
The cost of that performance.
Frequency and urgency: unknown.
Gains
The agent does the task without another prompt and without being pointed at the source.
He treated this as secondary, not the pain they lead with.
Two profiles, not one, once the buyer and the user are different people. In the first startups they may be the same person. The pitch uses the pain and the activities, not the feature list.
Business model
The shape was drawn. Almost none of it was filled.
The session’s point was that investors will ask these questions, and that a guess with no call behind it will show. Empty cells stay empty.
Piece
Version 0
Who
The segment above. One segment at a time.
Offer
The value sentence is still open.
Channel
Founder-led. Email versus LinkedIn not chosen.
Relationship
Not discussed. Who supports the first pilots is open.
Revenue
Not chosen. Design-partner terms are “hear more,” not a price.
Cost
Infrastructure is expensive enough that they are talking to angels. No cost sheet in this session.
People and work
Not listed.
Design partners
One company was named. Two are still to be picked.
Dynamic Mockups is the first candidate because he knows the people. It is not yet a fit. The public team looks smaller than the 4–5 engineer bar he set an hour earlier in the same session.
1. Dynamic Mockups — Luka Filipović, CEO
Said in the session He knows most of the people, so there is trust. The company has existed for more than a year, so the product is past a throwaway build. He believes they are AI-forward from what they do on LinkedIn. He does not know the engineer count and said he would ask. He assumes they have the problem. He knows the CEO, Luka. He does not think he knows the CTO. The first message goes to Luka.
Checked afterward Dynamic Mockups was founded in 2023 in Belgrade, by Luka Filipović and Miloš Medić. Their own help center calls Miloš the engineering lead. Stitch, launched late August 2026, is their embroidery digitizing agent. LinkedIn lists on the order of 9 people; the Serbian registry for 2025 lists 4 employees. Either figure can be wrong in a different direction. Both sit under a team of 4–5 engineers unless most of the company is engineering.
Why this call still happens. It tests the hypothesis he already named. If Luka says a team this size already knows who does what, and the product fits in one agent’s context, Dynamic Mockups is the falsifying case, not a failed outreach.
What to learn. Have they already tried to give an agent the company’s knowledge, and what was missing? Do they feel the performance and cost problem, or only the feature launch? How many people are actually engineering? Who, besides Luka, would use it?
What the call is not. A demo, a pilot offer, or proof that anyone will pay.
2 and 3. Not named
Said in the session He said he needed another day or two, possibly longer, to ask people about team structure before treating them as design partners. The session ended with one name.
The next two have to clear the same bar: a product company, product already in the world, about 4–5 or more engineers, AI already in the work, someone he can write to himself, and a reason to think the agent-context problem is theirs. An agency or an outsourcing team is out. A team that only said “cool” at a hackathon is out until they will talk about a pilot.
Outreach
Two messages to Luka. The product waits for the reply.
His draft led with Nucipal and with percentages he does not have. This follows the correction from the session: personal, then the problem, then a question. Nucipal appears only if he recognizes the problem.
First message
Zdravo Luka,
video sam launch Stitch-a. Svaka čast, ozbiljna stvar.
Vi ste produktna firma i proizvod vam već živi neko vreme. Kod timova koji tako grade, agent često nema znanje firme: model nije treniran na tome kako vi radite, a dogovor iz Slacka mu ne uđe u kontekst kad treba da odradi zadatak. Onda performanse i trošak agenta postanu problem.
Kako je to kod vas? Da li ste već pokušavali da agent vidi kako Dynamic Mockups zapravo radi, i šta je falilo?
Igor
Only if he says the problem is real
Hvala, to mi znači.
Pitam jer radim na Nucipalu. Hoću da agent ima to znanje firme bez ručnog održavanja dokumenata. Pre svega hoću da razumem kako vi tome pristupate.
Ako imaš 20 minuta ove ili sledeće nedelje, možemo kratko. Bez dema.
Uz to, koliko vas je trenutno u inženjeringu? Treba mi da vidim da li ste uopšte tim za koji ovo ima smisla.
Still open after the session
What Version 0 does not pretend to know.
The value sentence. The draft above is for him to accept or replace.
The second and third design partners.
Email or LinkedIn, and whose network goes first.
The smallest product that counts as ready for a pilot.
The number that counts as success on the first batch of messages.
How it is priced after a pilot.
The moat: what a better-funded team could not copy.
Whether Glean’s scale and contract size, as he stated them, still match public figures.
CliftonStrengths, the free personality tests, the Ominimo interview, and the sales-call video were left as optional. They did not change this board.