I’ve been watching an interview with the founder of Manus. Their journey is a powerful reminder that strong engineering and product execution can create their own momentum.
Today, from Ukraine, we’re opening a $1B+ strategic financing process for 1UA-AGE. 🧵
@mohsin_121 Exactly. This is why model capability and engineering authority must remain separate.
The agent may propose and execute, but it must not redefine the contract used to judge its own work. Any authorised change must be explicit, attributable and versioned—and must trigger...
OpenAI reports GPT‑6 Astra completing a PCB-layout demonstration in KiCad in 2:54.
Because the starting state, complexity, hardware, constraints, acceptance procedure and engineering files are undisclosed, 1UA‑AGE makes no normalised runtime comparison.
Status: internal reproducible results, prepared for independent review.
Not independently verified. Not physically validated. No claim of superiority.
A disclosure-safe evidence package will follow the first engineering review.
AI can generate. Engineering must prove.
While preparing the next PCB benchmark evidence for 1UA-AGE, we hit a different engineering problem:
how much authority should a high-agency AI have when it believes the human-defined condition is wrong?
The 1UA-AGE PCB benchmark evidence is still being completed.
We will publish it when the reproducibility gates are actually closed — not when we get one convenient result.
Discovery creates a proposal, not authority.
Verification is part of the product.