Technology Executive & Board Member. President & CPO, Cisco. Proud dad. Love design. Views are mostly my own, but sometimes not entirely my own ;-)
Silicon Valley
- Tweets 3,814
- Following 2,040
- Followers 26,783
- Likes 7,425
Pinned Tweet
There is such a profound shift occurring in the way that Agents alter our infrastructure requirements.
We are entering a networking supercycle. It’s not because humans are consuming more content. It’s because machines are beginning to think, act, and transact continuously.
Cisco's latest research on AI traffic patterns points to something much bigger than incremental bandwidth growth. Enterprise WAN traffic without agentic AI was projected to grow roughly 2.5x over the next
decade. With agentic AI, that projection jumps to ~9x.
And here’s the craziest part! After following this data closely, I believe even those numbers may prove to be wildly conservative. This is the first time when we have published a study like this where I feel that the projections might be off significantly and what we might think takes a decade happens in 3 years.
Why?
Because most people are still modeling AI like software. It is not. AI behaves more like a new species of digital labor. A SaaS app waits for humans. Agents do not. Agents continuously reason, retrieve, coordinate, negotiate, execute, and loop. At software
speed. Without pause. 7x24. They never get sick. Don’t need a vacation. Dont get tired. Don’t need sleep.
That creates a fundamentally different traffic architecture. The industry spent decades optimizing networks for bursty downloads, video streaming, and
human-paced interactions, almost all of it flowing downstream to a person on the other end.
AI traffic inverts that. A single agentic task can generate 450% more traffic than a human doing the same work.
Roughly 70% of that is inference. And nearly 10% of AI flows now carry more upstream than downstream data, versus 0.5% for typical web traffic, because context continuously moves back into models. Network traffic is not just increasing in bandwidth. It is fundamentally getting reshaped.
This last point matters most. The internet was built as a distribution system for content. AI is turning it into an active system for cognition. The path between agents and models is becoming the spinal cord of intelligence itself. When that path degrades, the agent degrades.
Networking stops being a passive transport layer and
becomes part of the intelligence stack. That changes everything about how we think about resiliency, observability, security, and capacity at the edge. We may be grossly underestimating what is coming. The future will not simply have more users online. It will have trillions of digital coworkers operating continuously on behalf of humans, enterprises, applications, and eventually physical systems.
Humans click.
Agents swarm.
That difference is what creates a supercycle. This supercycle of inference infrastructure will not just be compute bound, but also memory and network bound.
Take a look at the report here: cisco.com/c/dam/en/us/soluti…
23
29
203
75,536
A modern day case study of an absolutely EPIC marketing campaign is Muse. What a great freaking job @Meta has done not just in building a category creator but also just owning the space. Very cool to see!
5
1
44
4,335
So true at so many levels @jaltma 👇🏾
1
2
18
2,430
Thanks @amitisinvesting for having me on your pod. Great chat. Also thanks to @PalantirTech and Alex for including us in AIPCon.
Sat down with the President and Chief Product Officer at Cisco, Jeetu Patel, to discuss how agentic inference is increasing the need for security and networking, what challenges and opportunities Cisco sees for enterprise AI adoption, and how their partnership with Palantir will help enterprises get the strongest source of alpha from expanding usecases with open source models.
Thank you to Jeetu @jpatel41 from Cisco @Cisco for sitting down and taking the time to talk at Palantir's 11th AIPcon!
3
2
92
20,397
Really interesting analysis by @JayaGup10 👇🏾
If the markets took DeepSeek seriously as a threat to frontier labs, they should be taking Jev (and Jev like-models) far more seriously. Here is why:
1. It's built for every developer, not just ML teams.
Until now, cutting model costs meant open-weight models: choosing a base, hosting it, fine-tuning it, evaluating it, and retraining it. That's an ML team's job, and most companies don't rlly have one. Jev is an API - it works on day one with no GPUs and no training.
2. It's so cheap that everyone is trying it, and everyone is suddenly discovering the waste.
At a fraction of a cent per decision, there's no reason not to test it. Anyone can paste the API key into Claude or ChatGPT and ask it to find savings, including non-technical people (so many talent teams are already using it) Curiosity is turning into discovery: companies see how much they've been paying flagship prices for yes-or-no answers. Teams that never cared about inference cost now do, including non-tech companies.
3. It hits every term of the frontier revenue equation.
Frontier revenue is (calls) × (tokens per call) × (price per token). Decision / Jev models pressure all three.
a.) Fewer calls. Frontier valuations assume the labs capture the agent boom: millions of agents making billions of calls. But most agent steps are decisions: which tool, is this done / safe / escalate / classifying / tagging /routing. Simple generation moves to cheaper and open models. Retries, wasted steps, and escalations disappear !
b.) Smaller calls. Routers strip stale context and unneeded tools, so the calls that still reach the frontier cost less.
c.) Cheaper calls. Once routers show cost per task side by side, flagship pricing on routine work doesn't hold.
4. It puts a router in front of and within every AI product.
Routers existed before, but they were hard to build and unreliable. Decision models fix both. They're the ideal router: fast, nearly free, and they report how confident they are, so the router knows when to escalate. Routers are now becoming a built-in setting in the tools developers already use, so companies get one without building anything.
Today, the frontier model runs the agent: it decides what to do next and calls tools. With decision models, ordinary code and cheap decisions run the loop, and the frontier model becomes a function the software calls when it needs heavy reasoning
5.) Every defense the labs have speeds it up. Price cuts confirm customers were overpaying. Shipping their own routers teaches customers to route, and makes it easier to route to competitors! Cheap decision tiers move the labs' own customers onto cheaper products and cannibalize revenue.
6.) They make post-training economical, and simultaneously increase the demand for it. One of the higher costs in RL on real enterprise work has been grading tasks without a checkable answer. A cheap, calibrated grader removes that cost. Environments become cheaper to run, router logs show which expensive workloads are worth replacing, and every team that discovered its overspend becomes a curious buyer. Companies that train, serve, and retrain custom models get cheaper inputs and a pipeline of motivated customers at the same time.
If DeepSeek made the market question the supply side: the cost of pretraining, the durability of the model moat, and how long a frontier model stays frontier, decision models make it question the demand side: how many calls actually need a frontier model at all.....
4
1
16
6,369
The risk of underestimating the capacity for AI is far greater than the risk of overestimating the capacity requirements and being slightly off on timing at this point, especially given the personal agent movement that is being catalyzed.
4
1
32
1,619
In addition to the potency of the product, your @Muse brand building is pretty epic @alexandr_wang. Congrats to you and the entire team. You all are creating the second ChatGPT moment in AI.
3
1
40
5,944
We Are About To Experience The Personal Agent Supercycle
AI is rarely accused of being underestimated. But I think the movement forming around personal agents is highly underestimated.
We have been talking about agents for a while. But outside the AI-pilled corner of X, I don’t think most people have internalized what happens when anyone can have a capable coworker that keeps working after they close their laptop.
My thesis is that personal agents will trigger another compounding AI supercycle. Consumer adoption, enterprise demand and infrastructure consumption will accelerate together.
Consider the early signals. In August, @OpenRouter data shared by @a16z showed agents consuming nearly 5x as many tokens as human-driven usage on its platform. @Deloitte projects inference will account for roughly two-thirds of AI compute in 2026. One person delegating a goal can set off an extended chain of planning, execution and verification. The amount of work we delegate becomes a new driver of demand.
Now combine that with mass consumer distribution. @Meta’s @Muse brings personal agents into @WhatsApp. Meta knows how to make sophisticated technology accessible to enormous audiences. That is a meaningful advantage in a category where adoption depends on people experiencing something they struggle to understand from a description.
Whether it is @Muse, Instinct, @Grokbot, @OpenAI or another player, personal agents could make power-user capability available through an ordinary conversation. People won’t need to master a collection of tools to experience what those tools can accomplish.
My bet is that the next 12 to 15 months will bring a much faster move toward mainstream adoption than most companies are planning for.
And that experience will travel into the enterprise. Once people routinely delegate meaningful work in their personal lives, they will expect similar capabilities at work. Consumer adoption becomes a training ground for enterprise adoption. Companies will face growing pressure from their own employees to make this possible.
The biggest constraint, in my view, will be trusted delegation and infrastructure availability.
An agent’s usefulness expands with the access we give it. So does the consequence of a mistake. Without access to our applications, data and tools, much of its potential remains inaccessible. Granting that access requires confidence in its judgment, clear limits on its authority, visibility into its actions and the ability to intervene.
That makes safety and security central to adoption itself. Every improvement in trust can unlock more delegation. More delegation creates demand for the compute, memory, networking and power needed to support it.
This is the compounding effect I think we are underestimating. More people discovering what agents can do. More work entrusted to them. More enterprise demand. More infrastructure required.
We are approaching another ChatGPT moment in AI. I thought @OpenClaw might be that moment, but its setup complexity limited its reach. The next breakthrough may come from agents that hide that complexity entirely: a Muse, Instinct, Grokbot or OpenAI product that lets anyone delegate a long-running task in a simple conversation and return to a completed result.
Delegation will feel natural, trustworthy and reliable at scale over the next few months. Every task users are willing to hand over will expand the market for agents and the infrastructure required to run them. The next ChatGPT moment is fast arriving. The next AI supercycle is forming right in front of our eyes. 3 months from now will look very different for personal productivity provided safety and security make progress which they will, however we will continue to have demand outstrip supply in infra if personal agents proliferate as much as I think they will.
26
19
116
106,452
This is so true. Someone once told me that politics is when the same person says something entirely different in two meetings on the same topic depending on who is in the meeting.
This is a great way to eliminate politics.
Groq Founder @JonathanRoss321 shares the biggest leadership lesson he learned from Nvidia CEO Jensen Huang:
“There is no circumstance where Jensen has one-on-ones with people.”
“When you're leading groups of people, if you want to reduce the amount of politics, stop having one-on-ones. Have big meetings with everyone who you want to tell something to and tell them all at once.”
“Copy everyone on the email.”
“If someone says, ‘Hey this person is screwing up,’ copy that person on the email. Let them jump in. Otherwise you're allowing politics to happen.”
“There's no politics. It's the least political large organization you will ever see.”
1
2
55
20,443
This is so good and so true. For all the early in career people who want to make a $1B before you turn 25, read this. By all means have the aspiration but also know that durability of success matters more than a flash of luck for fulfillment.
Putting in the hours and finding joy in the mundane work when others are just focused on the outcome is such a unique differentiator if we can get our mental state to that place. Every time I have decided to not throw in the towel when shit got really hard are the times I have cherished most in my professional life.
Taking the same shot 10,000 times, 100,000 times, a million times makes it impossible for us not getting good at something. Thats how we are built. We can master anything given enough time, provided we put in the work. It is such an important lesson. And every time I personally get obsessed even at this age with the result over the process, this will act as a good reminder to focus on the process because reps matter. 👇🏾
This paragraph by Kobe Bryant hits hard:
“Everyone wants to be a beast. Everyone wants to be the best. But very few people are willing to do what it actually takes. Because what it takes is boring. It is waking up at 4:00 AM. It is shooting the same shot a thousand times. It is watching the film when you are tired. People fall in love with the result, but they hate the process. You have to fall in love with the boredom. You have to fall in love with the repetition. If you can find joy in the mundane work that no one else sees, the lights will eventually shine on you.”
5
6
89
19,601
The most valuable AI relationship will not be with the model you ask questions. It will be with the agent you trust enough to say, “Handle it.”
Try either of these. The last mile in some of these bits is finally nailed for consumer use cases.
@instictlabs @Muse @bot
Others will follow soon. But it’s truly a game changer. I’ve now been using these on and off 10 or so days and I’m hooked.
The big difference unlocks with agents that has been discussed in coding for a while but is now starting to make its way for everyday tasks is tools access, long running agents and the agent getting better and better to understand you. I also think that atleast as of now, the degree of mental clarity it gets me so that I can do better work because I have the cycles to think is non trivial.
And lastly, you feel like you have this agent that is always by your side to carry out any range of tasks. It’s truly a huge unlock.
The last thing to note is that yes the model is crucial, but the product sensibility is truly what gets it to unlock in very different ways. Judgement in product decisions is a non-trivial part of the success of these agents.
14
5
63
14,263
The debate this week around the risks of safety and security with AI was the strongest we've seen. I happened to be in London earlier today and met with @cnbcKaren, Steve Sedgwick and @Benboulos from @CNBC.
We discussed a range of issues including safety and security, margins for hyperscalers, durability of demand, slowing down AI, global coordination to handle security risks and more.
7
3
58
186,658