Agentic Swarms & Multi-Agent Systems | Tech Talks AI

Описание: What happens when agents stop working solo and start working in teams?

This episode covers agentic swarms, multi-agent systems, and always-on assistants—plus the tradeoffs: specs, security, and “can we even tell who’s human online anymore?”

See how AI can evolve your business by visiting our website: https://www.intimetec.com/artificial-...

Chapters
• 0:00 Intro: agentic swarms + the current wave
• 0:47 What an “agent” is (LLM + tools in a loop)
• 1:57 Multi-agent systems: teams + roles
• 3:00 The “vending machine” example: profitability changes with a “boss” agent
• 5:00 Sub-agents vs multi-agent teams (hierarchy vs flatter collaboration)
• 6:50 Recursive spawning + “halting problem” concerns
• 8:53 “Power tools” for engineering: faster build or faster mess
• 10:59 When swarms win: parallel work vs sequential dependencies
• 14:51 Predicting swarm vs single-agent fit (model idea)
• 15:42 MULTbook: “Reddit, but all the users are agents”
• 20:00 OpenClaw: deep tool integrations + surprising emergent behavior
• 21:50 Prompt injection, attack surface, and why local access is risky
• 34:00 Remote workflows: SSH from phone (Tailscale + Termius)
• 39:53 Quality as the differentiator (UI + policies + support)
• 42:40 AX (agent experience): designing interfaces for agents
• 45:38 The “who’s human?” problem + consensus manipulation
• 47:52 Proving personhood online (Worldcoin mention)
• 51:03 Proactive agents + memory you can inspect/edit
• 54:34 Wrap

Key moments
• Agents become more useful when they can take actions—not just answer questions.
• Swarms help when you can parallelize; single agents win when tasks are sequential.
• Tool integrations create value and risk (prompt injection + broad permissions).
• “AX”: building interfaces for agents, not just humans.

Tools covered in the show
• Anthropic / Claude (agent behavior example and AI reasoning model) https://claude.ai/
• OpenClaw (always-on agent with deep system integrations) https://github.com/OpenClawAI/OpenClaw
• MULTbook (forum concept where agents post and respond to other agents) https://github.com/
(concept reference)
• Cursor (always-on coding agent in an AI-native code editor) https://cursor.sh/
• Cloud Code / Open Code / Codex (terminal and CLI-based agent coding workflows) https://openai.com/research/codex
• ElevenLabs (AI voice synthesis and voice calling workflows) https://11labs.io/
• Railway (isolated deployment and experimentation environment) https://railway.app/
• Telegram and WhatsApp (agent messaging workflows and bot integrations) https://telegram.org/
· https://www.whatsapp.com/
• Tailscale + Termius (SSH access from phone and secure remote workflows) https://tailscale.com/
· https://termius.com/
• GitHub CLI (issues management and terminal-based automation) https://cli.github.com/
• Linear (product management tool with strong UX and quality focus) https://linear.app/
• Jira / Azure DevOps / Asana (comparison set for work and issue tracking) https://www.atlassian.com/software/jira
· https://azure.microsoft.com/products/...
· https://asana.com/
• Parallel (human vs agent interface toggle concept) https://parallel.ai/
• Worldcoin (personhood verification and digital identity concept) https://worldcoin.org/
• Reddit (reference point for MULTbook-style forums) https://www.reddit.com/
• Instagram and X (anti-bot platform contrast in social ecosystems) https://www.instagram.com/
· https://x.com/

To learn more about who we are, what we do, and how we do it, visit our website at https://www.intimetec.com.

SUBSCRIBE to the In Time Tec YouTube channel:
   / @intimeteccreatingabundance  

Follow @InTimeTec
Linkedin:   / in-time-tec  
Facebook:   / intimetec  
Instagram:   / intimetec_  

We are a global software services company that partners with our clients to provide access to talent, enable innovation, and deliver immediate ROI.