📌 What was built and why it failed 0:00
"I built an agentic platform that made zero return on investment. Most businesses seem to be following the same path I did and making the same mistakes."
The product was a deal-sourcing platform for private equity firms. A user entered an investment thesis in natural language — "we want mid-market manufacturing companies with EBITDA over $2M in the Midwest" — and the platform routed LLMs across company data sources, translated the thesis into a search, and returned matching businesses. It was a full self-service SaaS: payment integration, credits, teams, organisations. A few dozen users ran their real investment theses through it.
The product worked. "In essence, problem solved." It identified companies that matched the thesis. It returned nothing.
What checked out. Brainqub3 (
brainqub3.com) is a real company — HTTP 200. The presenter states he has been delivering custom software and agents for three and a half years. The platform description is specific: PitchBook on one side as the raw data provider, brokers on the other side as the human service layer, and his platform positioned in the middle as the automated discovery layer. The competitive landscape is recognisable to anyone in private equity.🔍 The gap between identification and acquisition 6:20
"What it didn't do — it didn't actually facilitate the customer buying that company. Ultimately, that's the gap, isn't it?"
The platform identified companies. It did not help anyone acquire one. Between identification and acquisition lies:
- Owner outreach. Somebody has to call the owner and ask whether they would entertain an offer. "This is a very human process. This is a negotiation."
- Due diligence. Inspecting the data room — the documents the seller provides for verification. Largely manual.
- Debt financing. If the buyer is raising debt, somebody has to build the case for the bank. "Our solution didn't solve that problem either."
The platform solved the part that was already solvable — searching databases for matching companies. PitchBook and its competitors already did this, just more slowly. The platform automated the search and left every other step untouched. The customer's actual goal — deploying capital to buy a business — sat on the far side of all of it.
⚡ The weekend that beat six months 14:44
"Over a weekend, I created a proof of concept. I just hooked Claude Code up with some Skills and allowed it to use some of the data sources. The feedback from the customer was — wow, these results are amazing. They're actually superior to the ones I got when I used the platform. In essentially a weekend, I had created something that was better than the platform I'd been creating for six months."
| 6-month platform | Weekend rebuild | |
|---|---|---|
| Architecture | Custom routing, custom orchestration, payment, credits, teams, SaaS | Claude Code + Skills + data source connectors |
| Customer experience | Self-service platform with login, credits, dashboard | Direct results sent to the customer — no platform at all |
| Customer feedback | Good, constructive. Used it. | "Superior to the platform." Asked for more. |
| Technical debt | Accumulated — bugs, customer queries, platform maintenance | None |
"The customer didn't give a damn about the platform. The customer cared that the businesses that were returned were businesses they liked. The platform was just getting in the way."
He built the platform before Claude Code and Codex existed. The custom orchestration — routing LLMs, stringing them together — became redundant the moment general-purpose agent harnesses shipped. "We built on the assumption that the technology wasn't really going to get better. We did a lot of custom things to try and string LLMs together, which now is pretty much redundant because of what's available out of the box."
That sentence is the architectural version of the outcome failure. He built for the technology that existed; the technology improved; his platform became redundant; a weekend with the new tools outperformed it.
That sentence is the architectural version of the outcome failure. He built for the technology that existed; the technology improved; his platform became redundant; a weekend with the new tools outperformed it.
📋 The three lessons
1. Service first, platform second 12:00
"We wouldn't have gone platform first. We would have gone service first. A service gives you the flexibility to work around the customer while you learn the real journey to the outcome. A platform locks the customer into a way of working nobody has proven yet, and it accrues technical debt that never earns its keep."
2. Aim at the true outcome, not the partial one 11:00
"Retrieving information from a database — if you're building agentic AI to serve businesses, that outcome is too low a bar now. You need to shoot for more ambitious outcomes to be taken seriously by customers."
3. ROI does not come from better agents 22:26
"Return on investment doesn't come from better agents. It's not going to come from a stronger model. It's not going to come from Fable 15 or ChatGPT Galaxy 1000. It comes from understanding clearly the outcome you're trying to get from your customers, and deploying agents at the high leverage points to make that outcome occur faster, cheaper, more accurately than your competitors."
👤 The human problems 18:18
"Even after doing all of that — identifying the businesses, picking up signals from the internet — somebody still has to get on the phone to the business. Somebody still has to talk to the owner. Somebody still has to build a relationship. I can't imagine too many business owners accepting an offer from a company that simply asked an agent to contact them."
His distinction is clean: some problems are not agent-shaped. Calling a business owner and asking whether they want to sell is not an automation problem — it is a relationship problem. Agents can find the candidates. Agents cannot close the deal. The skill is knowing which is which, and deploying agents only where they add leverage.
The inversion at the end is the durable part: "Not everybody can deliver the outcome. Not everybody can deliver the service that is faster, more efficient, more precise and cheaper than the competitor — because your service leverages agentic AI in the right places. And your service knows when to delegate to humans and when to delegate to agents. That's the real opportunity."
The agents are the leverage. The service is the product. The platform was an expensive distraction.
The agents are the leverage. The service is the product. The platform was an expensive distraction.
🔍 Claims checked
| Claim | Result |
|---|---|
| Brainqub3 is a real company | brainqub3.com HTTP 200 |
| The presenter has delivered custom software/agents for 3.5 years | His claim; consistent with the specificity of the platform description |
| Six months of development, full SaaS platform with payments | His claim; the description is internally consistent and technically specific |
| PitchBook is a real PE data provider | pitchbook.com exists — well-known in private equity |
| Weekend rebuild with Claude Code + Skills outperformed the platform | His claim; consistent with the demonstrated capabilities of Claude Code + Skills |
| Customer feedback: weekend results were "superior to the platform" | His report of customer feedback; no independent verification |
Method note. YouTube caption routes returned HTTP 403, so the audio was downloaded with
yt-dlp and transcribed locally with Whisper large-v3-turbo — 301 segments across the full 23:29. It rendered Brainqub3 as "BrainCube", PitchBook as "PitchBook", Claude Code as "Claude Code" and "CloudCode", Codex as "Codex" — all confirmed against the company's website and the description. The company's website was verified; the platform claims are his. Verified 5 August 2026.💡 Key takeaways
- A weekend with Claude Code + Skills beat six months of custom platform development. Not metaphorically — the customer said the weekend results were superior, and the platform was abandoned.
- Outcomes are king; software value has collapsed. "Everybody now believes they can build software with Claude. The perceived value of software has decreased." Retrieving information from a database is too low a bar.
- Service first, platform second. A platform locks you into an unproven way of working and accrues technical debt. A service lets you learn the real journey to the outcome before automating it.
- Some problems are not agent-shaped. Calling a business owner, building a relationship, closing a deal — these are human problems. Agents can find the candidates; they cannot close.
- ROI does not come from better models. It comes from understanding the outcome and deploying agents at the highest leverage points on the way to it.
- Don't build for the technology that exists — it will improve. Six months of custom orchestration became redundant the moment Claude Code shipped.
🔗 Resources & links
- Watch This Before You Pay Anyone to Build You an AI AgentSource video — Brainqub3, 23:29, 5 August 2026
- Brainqub3The presenter's company — Always-On Agent Blueprint and Claude Skills Playbook
🕐 Timestamp index
0:00Zero return on investment
1:48The problem — PE deal sourcing
4:00How it was built — custom LLM routing
5:22Six months, full SaaS, users on the platform
6:20What it didn't do — the gap
7:14Owner outreach, due diligence, debt financing
9:40Outcomes are king — software value has collapsed
12:00Lesson 1: Scrap the platform, go service first
14:44A weekend with Claude Code beat six months
15:51"The customer didn't give a damn about the platform"
17:07Lesson 2: Pick more realistic outcomes
18:18Somebody still has to get on the phone
22:26ROI doesn't come from better agents