Aminreza Khoshbahar

Chief engineer, Genesis Ventures. One operator + agents closing small turbojets and airframes against fixed briefs. Orion → T1.

[Aminreza Khoshbahar @RealAminreza]

I pointed my agent system at the physical world. It designed a jet engine concept end to end — closed the cycle itself, generated 561 B-rep solids, then checked its own work: 59 of 59 validation checks!
13.34 kN thrust. 165 kg dry. 21,000 rpm. Every figure published. more to come!!

[Aminreza Khoshbahar @RealAminreza]

@zaidmukaddam Cope.

[Aminreza Khoshbahar @RealAminreza]

the buckling line is the one that should stick, because it isn't a taste thing, it's a default. ansys topology optimization won't even accept an eigenvalue buckling system upstream — you get static structural, modal, harmonic, thermal, and it auto-inserts compliance as the objective with a mass constraint. fusion generative has a buckling safety factor but it's marked optional, step 7 of 8, so it's off unless you went looking for it, and lateral-torsional isn't a criterion in any of them. sigmund's group has been publishing on this since 2019, compliance minimisation converges on slender members in pure compression which is the exact geometry that goes unstable, and it gets dropped because the eigenvalue problem is expensive. a 50% mass number off one static load case is answering the question that wasn't binding.

[Aminreza Khoshbahar @RealAminreza]

The dangerous misconception is thinking “agent + CAD = engineering.” MCP can generate geometry; it doesn’t magically close requirements, load cases, thermals, CFD/FEA, materials, tolerances, manufacturability, failure modes, test and validation. Until the agent is inside that verification loop, you haven’t automated engineering , you’ve automated drawing things that look engineered.

[Aminreza Khoshbahar @RealAminreza]

@rezoundous Then comes Fable 6.

[Aminreza Khoshbahar @RealAminreza]

@ArtificialAnlys @its_all_go you were saying.

[Aminreza Khoshbahar @RealAminreza]

@signalgaining Bro..

[Aminreza Khoshbahar @RealAminreza]

@its_all_go @ConnorTalksAI OpenAI fanboy confirmed.

[Aminreza Khoshbahar @RealAminreza]

@its_all_go @ConnorTalksAI alright buddy.

[Aminreza Khoshbahar @RealAminreza]

@its_all_go @ConnorTalksAI excuse me what?

[Aminreza Khoshbahar @RealAminreza]

@ibuildthecloud did you built the sheepherdr?

[Aminreza Khoshbahar @RealAminreza]

I do believe in miracles.

[Aminreza Khoshbahar @RealAminreza]

@ibuildthecloud I like your vision

[Aminreza Khoshbahar @RealAminreza]

@bot Why not release the beta for Linux instead of MacOS???

[Aminreza Khoshbahar @RealAminreza]

"But, but but Opus 5 is BENCHMAXED!"

[leo 🐾 @synthwavedd]

Terminal-Bench 3.0 has been released! Claude Opus 5 Max takes the crown from GPT-5.6 Sol, achieving a SoTA score of 43.5% to Sol's 34.6%, with Fable 5 in 3rd within Sol's margin of error (34.1%).

[Aminreza Khoshbahar @RealAminreza]

Nobody denied the fact that OpenAI' latest line up is decent and in fact I would say they are enough for majority of daily tasks for most people! That being said, this is also true for Anthropic models! We are not discussing about a daily driver, anyone with half a decent serious experience in real engineering tasks would agree that Sol is not in the same league of Fable 5. Is it clear?

[Aminreza Khoshbahar @RealAminreza]

@TFAwinner @araseb_ OpenAI fanboy?

[Aminreza Khoshbahar @RealAminreza]

@KrisTalksAI There is zero doubt that local (consumer grade) and OSS models are absolutely necessary to push the frontier and to make intelligence available for everyone. However, regulation and safety for OSS can not be ignored at all.

[Aminreza Khoshbahar @RealAminreza]

@danellisona I guess because I can't dance in front of a jet engine?

[Aminreza Khoshbahar @RealAminreza]

@OpenAI FINALLY!!!

[Aminreza Khoshbahar @RealAminreza]

@LocalAiCherry give it a try

[Aminreza Khoshbahar @RealAminreza]

Facts.

[Deniza @dmalataeva]

ok hear me out:
Playing Minecraft teaches you to be a well-rounded tech lead