StarZero All shows Try the podcast clipper →

Apr 13, 2026 · 33m · big-technology

How AI Agent Swarms Might Be AI's Next Leap — With Guillaume de Saint-Marc

Guillaume de Saint-Marc · 25m spoken Alex Kantrowitz · 4m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

Guillaume de Saint-Marc, VP of Engineering at Cisco Outshift, discusses the technological evolution from isolated AI models to distributed, self-forming multi-agent swarms powered by open networking standards, semantic coordination layers, and robust security controls. He explains how frameworks like the Internet of Agents and Task-Based Access Control enable scalable, secure enterprise collaboration across heterogeneous AI ecosystems.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 13.4% of the talking time here. How this is scored →

Alex as informed peer 4.2 Guest teaching 5.3 Guest disagreement 0.3 Alex pushing back 1.5
05100:0010:0020:0030:003:46–7:36 · Alex as informed peer 3/10 Horizontal Scaling and the Internet of Agents Foundation Alex prompts the discussion by bringing up Moltbook as an example of agent networks. Guillaume explains Cisco Outshift's thesis on horizontal scaling versus monolithic models, detailing the Internet of Agents infrastructure and the limitations of deterministic workflows.7:36–15:42 · Alex as informed peer 4/10 The Mechanics of Self-Forming Agent Collaboration Alex pushes back by asking how self-forming agent swarms can be controlled to avoid runaway AI. Guillaume details architectural guardrails including network segmentation, MLS encryption, and task-based access control (TBAC) with semantic verification.15:42–18:36 · Alex as informed peer 3/10 Extending the OSI Model: Syntactic and Semantic Agent Layers Guillaume walks through Cisco's technical proposal extending the 7-layer OSI model to include Layer 8 (syntactic connectivity) and Layer 9 (semantic intent and cognitive coordination). Alex listens attentively and clarifies the spelling of the AGNTCY project.18:36–23:23 · Alex as informed peer 4/10 Real-World Application: Multi-Agent IT Outage Resolution Alex asks for a practical enterprise use case, and Guillaume explains multi-agent IT outage response coordinating SRE, security, and observability agents. Alex shows domain familiarity by defining site reliability engineering (SRE) for the audience.23:23–27:14 · Alex as informed peer 6/10 Heterogeneous Swarms, Specialized SLMs, and Cost Control Alex demonstrates strong technical and economic intuition, asking if swarms use specialized small language models (SLMs) for specific micro-tasks to manage inference costs. Guillaume validates Alex's reasoning and explains their TBAC SLM implementation.27:14–31:31 · Alex as informed peer 5/10 The Open-Source Philosophy and the Internet of Cognition Alex challenges Guillaume on why Cisco open sources key intellectual property instead of maintaining a proprietary competitive moat. Guillaume explains that open interoperability expands the market and that Cisco competes on innovation velocity.3:46–7:36 · Guest teaching 5/10 Horizontal Scaling and the Internet of Agents Foundation Alex prompts the discussion by bringing up Moltbook as an example of agent networks. Guillaume explains Cisco Outshift's thesis on horizontal scaling versus monolithic models, detailing the Internet of Agents infrastructure and the limitations of deterministic workflows.7:36–15:42 · Guest teaching 6/10 The Mechanics of Self-Forming Agent Collaboration Alex pushes back by asking how self-forming agent swarms can be controlled to avoid runaway AI. Guillaume details architectural guardrails including network segmentation, MLS encryption, and task-based access control (TBAC) with semantic verification.15:42–18:36 · Guest teaching 6/10 Extending the OSI Model: Syntactic and Semantic Agent Layers Guillaume walks through Cisco's technical proposal extending the 7-layer OSI model to include Layer 8 (syntactic connectivity) and Layer 9 (semantic intent and cognitive coordination). Alex listens attentively and clarifies the spelling of the AGNTCY project.18:36–23:23 · Guest teaching 6/10 Real-World Application: Multi-Agent IT Outage Resolution Alex asks for a practical enterprise use case, and Guillaume explains multi-agent IT outage response coordinating SRE, security, and observability agents. Alex shows domain familiarity by defining site reliability engineering (SRE) for the audience.23:23–27:14 · Guest teaching 5/10 Heterogeneous Swarms, Specialized SLMs, and Cost Control Alex demonstrates strong technical and economic intuition, asking if swarms use specialized small language models (SLMs) for specific micro-tasks to manage inference costs. Guillaume validates Alex's reasoning and explains their TBAC SLM implementation.27:14–31:31 · Guest teaching 4/10 The Open-Source Philosophy and the Internet of Cognition Alex challenges Guillaume on why Cisco open sources key intellectual property instead of maintaining a proprietary competitive moat. Guillaume explains that open interoperability expands the market and that Cisco competes on innovation velocity.3:46–7:36 · Guest disagreement 0/10 Horizontal Scaling and the Internet of Agents Foundation Alex prompts the discussion by bringing up Moltbook as an example of agent networks. Guillaume explains Cisco Outshift's thesis on horizontal scaling versus monolithic models, detailing the Internet of Agents infrastructure and the limitations of deterministic workflows.7:36–15:42 · Guest disagreement 1/10 The Mechanics of Self-Forming Agent Collaboration Alex pushes back by asking how self-forming agent swarms can be controlled to avoid runaway AI. Guillaume details architectural guardrails including network segmentation, MLS encryption, and task-based access control (TBAC) with semantic verification.15:42–18:36 · Guest disagreement 0/10 Extending the OSI Model: Syntactic and Semantic Agent Layers Guillaume walks through Cisco's technical proposal extending the 7-layer OSI model to include Layer 8 (syntactic connectivity) and Layer 9 (semantic intent and cognitive coordination). Alex listens attentively and clarifies the spelling of the AGNTCY project.18:36–23:23 · Guest disagreement 0/10 Real-World Application: Multi-Agent IT Outage Resolution Alex asks for a practical enterprise use case, and Guillaume explains multi-agent IT outage response coordinating SRE, security, and observability agents. Alex shows domain familiarity by defining site reliability engineering (SRE) for the audience.23:23–27:14 · Guest disagreement 0/10 Heterogeneous Swarms, Specialized SLMs, and Cost Control Alex demonstrates strong technical and economic intuition, asking if swarms use specialized small language models (SLMs) for specific micro-tasks to manage inference costs. Guillaume validates Alex's reasoning and explains their TBAC SLM implementation.27:14–31:31 · Guest disagreement 1/10 The Open-Source Philosophy and the Internet of Cognition Alex challenges Guillaume on why Cisco open sources key intellectual property instead of maintaining a proprietary competitive moat. Guillaume explains that open interoperability expands the market and that Cisco competes on innovation velocity.3:46–7:36 · Alex pushing back 0/10 Horizontal Scaling and the Internet of Agents Foundation Alex prompts the discussion by bringing up Moltbook as an example of agent networks. Guillaume explains Cisco Outshift's thesis on horizontal scaling versus monolithic models, detailing the Internet of Agents infrastructure and the limitations of deterministic workflows.7:36–15:42 · Alex pushing back 3/10 The Mechanics of Self-Forming Agent Collaboration Alex pushes back by asking how self-forming agent swarms can be controlled to avoid runaway AI. Guillaume details architectural guardrails including network segmentation, MLS encryption, and task-based access control (TBAC) with semantic verification.15:42–18:36 · Alex pushing back 0/10 Extending the OSI Model: Syntactic and Semantic Agent Layers Guillaume walks through Cisco's technical proposal extending the 7-layer OSI model to include Layer 8 (syntactic connectivity) and Layer 9 (semantic intent and cognitive coordination). Alex listens attentively and clarifies the spelling of the AGNTCY project.18:36–23:23 · Alex pushing back 1/10 Real-World Application: Multi-Agent IT Outage Resolution Alex asks for a practical enterprise use case, and Guillaume explains multi-agent IT outage response coordinating SRE, security, and observability agents. Alex shows domain familiarity by defining site reliability engineering (SRE) for the audience.23:23–27:14 · Alex pushing back 1/10 Heterogeneous Swarms, Specialized SLMs, and Cost Control Alex demonstrates strong technical and economic intuition, asking if swarms use specialized small language models (SLMs) for specific micro-tasks to manage inference costs. Guillaume validates Alex's reasoning and explains their TBAC SLM implementation.27:14–31:31 · Alex pushing back 4/10 The Open-Source Philosophy and the Internet of Cognition Alex challenges Guillaume on why Cisco open sources key intellectual property instead of maintaining a proprietary competitive moat. Guillaume explains that open interoperability expands the market and that Cisco competes on innovation velocity.

speaking balance: gold is Alex, purple is the guest (3 minute bins)

0:00 · Alex 29.4% · guest 70.6%0:00 · Alex 29.4% · guest 70.6%3:00 · Alex 22.9% · guest 77.1%3:00 · Alex 22.9% · guest 77.1%6:00 · Alex 2.6% · guest 97.4%6:00 · Alex 2.6% · guest 97.4%9:00 · Alex 14% · guest 86%9:00 · Alex 14% · guest 86%12:00 · Alex 11.2% · guest 88.8%12:00 · Alex 11.2% · guest 88.8%15:00 · Alex 0.1% · guest 99.9%15:00 · Alex 0.1% · guest 99.9%18:00 · Alex 18.6% · guest 81.4%18:00 · Alex 18.6% · guest 81.4%21:00 · Alex 13.5% · guest 86.5%21:00 · Alex 13.5% · guest 86.5%24:00 · Alex 11% · guest 89%24:00 · Alex 11% · guest 89%27:00 · Alex 15.5% · guest 84.5%27:00 · Alex 15.5% · guest 84.5%30:00 · Alex 9% · guest 91%30:00 · Alex 9% · guest 91%33:00 · Alex 0% · guest 0%33:00 · Alex 0% · guest 0%
Sharpest disagreement ▶ 27:39 Guillaume rejects walled-garden AI architectures

Guillaume firmly dismisses the proprietary walled-garden model, arguing that agentic intelligence must mirror the open internet to succeed.

 Hardest push from Alex ▶ 9:55 Alex presses on runaway AI risks

Alex interrupts to challenge the premise of autonomous self-forming agent swarms, asking how enterprise control can be maintained against runaway systems.

 Biggest teaching moment ▶ 16:20 Guillaume breaks down OSI Layer 8 and Layer 9

Guillaume explains how traditional networking models stop at Layer 7, educating the host on why multi-agent cognition requires new syntactic and semantic layers.

 Alex holds their own ▶ 25:21 Alex deduces hierarchical SLM cost structure

Alex accurately deduces that production agent systems must pair large orchestrator models with specialized SLMs to prevent runaway compute costs.
the scores for every segment, with the reasoning behind each
ChapterTopicAlex as informed peerGuest teachingGuest disagreementAlex pushing backWhy
Horizontal Scaling and the Internet of Agents Foundation 3500 Alex prompts the discussion by bringing up Moltbook as an example of agent networks. Guillaume explains Cisco Outshift's thesis on horizontal scaling versus monolithic models, detailing the Internet of Agents infrastructure and the limitations of deterministic workflows.
The Mechanics of Self-Forming Agent Collaboration 4613 Alex pushes back by asking how self-forming agent swarms can be controlled to avoid runaway AI. Guillaume details architectural guardrails including network segmentation, MLS encryption, and task-based access control (TBAC) with semantic verification.
Extending the OSI Model: Syntactic and Semantic Agent Layers 3600 Guillaume walks through Cisco's technical proposal extending the 7-layer OSI model to include Layer 8 (syntactic connectivity) and Layer 9 (semantic intent and cognitive coordination). Alex listens attentively and clarifies the spelling of the AGNTCY project.
Real-World Application: Multi-Agent IT Outage Resolution 4601 Alex asks for a practical enterprise use case, and Guillaume explains multi-agent IT outage response coordinating SRE, security, and observability agents. Alex shows domain familiarity by defining site reliability engineering (SRE) for the audience.
Heterogeneous Swarms, Specialized SLMs, and Cost Control 6501 Alex demonstrates strong technical and economic intuition, asking if swarms use specialized small language models (SLMs) for specific micro-tasks to manage inference costs. Guillaume validates Alex's reasoning and explains their TBAC SLM implementation.
The Open-Source Philosophy and the Internet of Cognition 5414 Alex challenges Guillaume on why Cisco open sources key intellectual property instead of maintaining a proprietary competitive moat. Guillaume explains that open interoperability expands the market and that Cisco competes on innovation velocity.

Statements from this episode (12)

Prediction Not checkable as stated
Saint-Marc: Superintelligence will emerge from multi-agent scaling, not single models
“We had this strong thesis and vision that the intelligence will come not just from building ever smarter and ever bigger and more powerful single agents But more through what we call horizontal scaling. So we distributed systems, a lot of agents coming togethe…”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 3:54
Assertion Supported
Saint-Marc: Cisco created Linux Foundation project Agency with Google, Oracle, Dell
“To back all this effort, we've actually created an open source project called Agency. And this, you know, became a Linux Foundation project with the backing of a few large formative members. Google, Oracle Red Hat were with us and Dell as well.”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 5:34
Insight
De Saint-Marc: Complex enterprise AI missions require self-forming agent reasoning systems
“Workflows are great because they are very deterministic. They can be reassuring at the beginning for enterprise because they can be certified. But if you really want to go after interesting mission, you need this self-forming reasoning aspect where actually at…”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 9:08
Assertion Supported
Saint-Marc: Cisco Outshift collaborates with Google on agent identity standards
“We work also very closely with Google on this and the A to A group. We are very active here.”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 12:54
Insight
Saint-Marc: AI agents without agency are useless, but unconstrained agency is dangerous
“An agent with no agency is useless. An agent with you know, too much agency can be dangerous and especially if it goes out of control.”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 14:56
Insight
Saint-Marc: Agentic AI requires adding Layers 8 and 9 to OSI model
“We've also published a paper extending the Aussie model. So, like this, you know, most engineers will know the seven layers of the Aussie model, the IP stack and the different communication layers, and level seven is really where the world has been living for …”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 16:25
Disclosure
Saint-Marc: Cisco open-sourced a multi-agent SRE system called CAIPE
“We entirely, so in my team, we've actually built our own agentic SRE system. It's a multi-agent system. It's called CAPE, C-A-I-P-E and we've entirely open sourced it with from with a group called Canoe, C-N-O-E”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 20:16
Assertion Not checkable as stated
Saint-Marc: Agentic root cause analysis cut troubleshooting from days to minutes
“Another agent will have, will be like the agent capable of finding troubleshooting, like finding what we call the root cause analysis, you know, doing the root cause analysis of the problem. We had some amazing results on this, literally going from putting exp…”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 22:08
Prediction Not checkable as stated
Saint-Marc: Future Enterprise Agent Swarms Will Be Deeply Heterogeneous
“As complexity is growing, and as we, again, as I said before, we asymptotically start to progress towards something we can call superintelligence, It's going to be very heterogeneous, very different types of agents, different vendors, different clouds, differe…”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 24:58
Disclosure
Saint-Marc: Cisco Outshift is developing an SLM for Semantic TBAC
“This is stuff that we are going to share soon, but like for the T-back functionality, the semantic teabag. For now, we are using you know, pick your model. So, connect the model you want to actually power the solution. But we are also working on a small langua…”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 25:43
Disclosure
Saint-Marc: Cisco Outshift expects over 80% of agent tech open-sourced
“Probably more than 80% of what we do needs to be open source because we actually believe in this importance of open and interoperable.”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 29:13
Assertion Supported
Saint-Marc: Four core Internet of Agents technologies are live today
“Observability, the connectivity, the identity, and the discovery of agents. All of these technologies are usable today and out there.”
Guillaume de Saint-Marc Apr 13, 2026 ▶ 29:55
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 300 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more