Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
6 contributions to AI Agents CLI | AI Automations
Start Here: Bring Your First Build to Life
This is the best place to start if you want to stop collecting AI content and start building. In this lesson we bring the environment to life: - terminal setup - Gemini onboarding - first simple Node-RED success After watching, answer this: What is the first system you want to bring to life in your own lab?
1 like • Aug 23
I’d bring a document-triage system to life first: watch one intake folder, classify invoices, extract a small JSON schema, and route low-confidence cases to a human review queue. Starting with one document type and keeping the original file plus model output makes evaluation straightforward, while the review decisions become useful feedback before expanding to more workflows.
Upgrade the Stack: From Basic to Real Workspace
This lesson is where the environment stops feeling basic and starts becoming a real operator workspace. We switch from Gemini to Codex, connect Dropbox, and make the agent more useful inside the working context. After watching, tell me this: What matters more for your work right now — stronger coding, native tools, or a better multi-agent flow?
0 likes • Aug 23
Native tools matter most for me because they remove the brittle glue between an agent and the workspace it must actually operate in. I’d start with a tightly scoped Dropbox tool—list, read, and write with explicit folder permissions—then log every call and returned error; that usually creates more dependable leverage than adding another agent before the basic tool boundary is trustworthy.
One Request -> First Working Telegram System
This is the first full business-use-case build in the lab. The important part is not just that a Telegram pipeline works at the end. The important part is that the AI agent builds the first working version inside a real environment, and the same system can keep evolving afterwards in the same context. After watching, answer this: If you had this workflow in your own environment, what would you change first — sources, formatting, approval logic, or delivery path?
0 likes • Aug 23
I’d change the approval logic first: every generated item should stay in a draft state until an explicit approve or reject action, with the decision stored against the job ID. Add an approval timeout and idempotent callbacks so a repeated Telegram tap cannot publish twice or rerun generation; once that boundary is reliable, sources, formatting, and delivery can evolve behind it.
Rebuild the OpenClaw video workflow
This course shows the full OpenClaw path from one Telegram request to a finished AI video. Watch the build, then try to map the workflow you would reproduce first in your own setup. After watching, answer one question: Which part would you rebuild first — the Telegram control flow, the Veo 3.1 step, the D-ID step, or the free webterminal practice setup at https://sell.systems/mod/assign/view.php?id=215
0 likes • Aug 23
I’d rebuild the Telegram control flow first, because it is the durable boundary around every expensive video step. I’d make each request a job with an idempotency key, persisted status, and explicit retry boundaries before wiring in Veo or D-ID; then those generators can be swapped without losing the user-facing state or sending duplicate videos.
Field Note: Why Physical AI Is an Operational Stack
I recently visited a robotics event in Hong Kong to observe physical AI outside the usual promotional context. Two test environments stood out. The first was a retail simulation where a mobile robot maps shelves and identifies where products are located. The second was robot soccer, which creates a compact test of movement, balance, coordination, timing and recovery. My three main observations: 1. Physical AI is a systems problem. A capable model is only one layer. The robot also needs reliable perception, localization, control, actuators, power and mechanical stability. 2. Imperfect demonstrations are highly valuable. When a robot loses balance or requires operator assistance, it exposes the real engineering bottlenecks instead of hiding them. 3. Human operations remain part of the product. Calibration, maintenance, recovery and supervision are still important parts of many real-world deployments. The result is not simply “AI inside a robot.” It is an operational stack that combines intelligence, hardware and ongoing support. Which physical-AI use case do you believe has the clearest commercial path today?
Field Note: Why Physical AI Is an Operational Stack
2 likes • Aug 23
Retail shelf mapping has the clearest near-term commercial path because the ROI is measurable: fewer stockouts, better planogram compliance, and less manual auditing. I’d start with a fixed route and human review for low-confidence detections, then expand autonomy only after comparing those metrics against the current process; that keeps calibration and recovery costs visible instead of hiding them in a demo.
1-6 of 6
Yousef H
1
2 points to level up
@yousef-halel-2063
Co-Founder at Vendo (YC S26). UC Berkeley honors grad at 19, Stanford MS. Building software that lets companies agentify products vendo.run is the web

Active 26d ago
Joined Aug 18, 2026