Founder @trybeniai backed by @aforeVC @untamedvc | @fdotinc Artifact alunmi | 9+ years building consumer products people use every day 💻
United States
Joined June 2026
- Tweets3.2K
- Following137
- Followers497
- Likes11.5K
Most consumer AI agents are either character chat or personal assistants that do things. The character chat feels niche, the personal assistant feels prosumer-only. Who's actually building for regular people like my mom? I don't need another agent to deal with my email or calendar.
Summer retweeted
Little preview of something I've been working on!
You can connect to any provider, or through any harness like Hermes Agent, and the setup is instant with no configuration required, unless you want to.
Proactive, self-aware, and will be fully open source
When you let users customize their AI companion, do they tweak it obsessively or just accept the default and move on?
Watching startups launch multimodal companions that do text, image, video, and audio equally well.
Nobody asked for all four. Everyone wants one thing done fast.
Summer retweeted
We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent.
It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.
My goat friend made my life building @trybeniai a bit better. If you use image to video, video model or creator making contents, check this out👀
Im gonna switch to Sume API for upcoming products and internal tools as well
As a builder in the AI video space, I was terrified about how much markup API providers charge for simply routing the models.
Higgsfield API adds a 40% markup even with its maximum advertised discount. Fal adds 100%. Runway adds 414% (Seedance 2 480p)
I might be hated by the industry, but :
Today, we’re launching Sume API, the first image & video model provider with 0% markup.
For example with Seedance 2.5 1080p
• 50% cheaper than fal
• 50% cheaper than higgsfield api (29% even with biggest discount)
• 28.02% cheaper than kie ai
The routing infrastructure layer is never the moat. We believe that APIs should be commoditized, and the moats will be built at the application layer.
Check the pricing comparisons below, and see how the broken industry is saved.
Create a high-quality, photorealistic 3D character reference sheet of my cat using the attached reference photos.
Identity Accuracy
Carefully analyze the photos and preserve my cat’s exact appearance, including facial structure, eye color, nose shape, ear shape, fur colors, markings, body proportions, tail, and unique characteristics.
All views must depict the exact same cat with consistent features and markings.
Layout
Create a professional 3D character reference sheet arranged in three horizontal rows on a clean, light gray studio background.
Row 1: Full Body Turnaround
Front
Front 3/4
Left Profile
Back
Right Profile
Sitting Pose
Row 2: Face & Body Details
Face Front
Face 3/4
Face Side
Top View
Bottom View
Front Paws
Back Paws
Row 3: Poses & Size
Lying Down
Loaf Pose
Reclining Pose
Size Comparison (with a simple gray human silhouette)
Visual Style
Premium photorealistic 3D rendering
Soft, realistic, detailed fur
Natural facial expressions
Accurate anatomy and proportions
Soft, professional studio lighting
Subtle realistic shadows
Neutral light gray background
Clean, symmetrical grid layout
The more I work with AI agents, the more I realize the bottleneck isn’t always the model. It’s the system around it.
I’ve been pushing agentic workflows pretty hard, experimenting with task decomposition, context management, tool orchestration, and different levels of autonomy. And one thing keeps coming up: throwing more tokens or reasoning effort at a problem doesn’t necessarily produce better results.
When an agent underperforms, I find myself looking less at the prompt and more at the architecture of the workflow. Is the task scoped correctly? Does it have the right context and tools? Are there clear checkpoints, validation criteria, and recovery paths?
The interesting part is that better models don’t automatically fix poorly designed workflows. Sometimes a simpler, more deterministic approach outperforms an elaborate multi-agent setup.
AI is making execution dramatically cheaper, but designing reliable systems is still hard. And I suspect that’s where a lot of the real leverage is.
the moment you ship something real to actual people beats the moment you polish something perfect for nobody. momentum > perfection.
watching AI tool launches like pokemon collecting at this point.
each one solving the exact same problem as the last 46.
what actually makes you switch?🤨