Cisco bets 2026 will be the year agentic AI hits enterprise scale
Cisco's CEO Chuck Robbins opened the company's AI Summit with a clear thesis: 2026 is the turning point for enterprise AI. Expect agentic applications to move from demos to deployed systems - with trust and security as the deciding factors.
"There are lots of questions and discussions about what does it mean to your enterprise infrastructure? What does it mean to your security posture? What does it mean to application development cycles?" Robbins said. His core concern: trust in data, models, infrastructure, agents, and partners.
The three constraints slowing enterprise AI
- Infrastructure limits. "We just don't have enough power, compute and network bandwidth," said Jeetu Patel, Cisco's president and chief product officer. Cisco is investing billions to build out AI infrastructure. He highlighted the P200 chip and the Cisco 8223 routing system for scale-out clusters that span multiple data centers - hundreds of kilometers apart - operating like a single ultra-cluster with deep buffering and coherent behavior. Patel also noted we're hitting physical limits of copper and traditional optics, making coherent optics a priority for data center growth.
- Trust deficit. "This is the first time that security is actually becoming a prerequisite for adoption," Patel said. Without trust, users won't deploy. That means assurance for data handling, model integrity, infrastructure security, and the behavior of autonomous agents - not just using AI for cyber defense, but trusting AI itself.
- Data gap. Public human-generated data is flattening out. "We're running out," Patel said. Synthetic data is becoming highly effective for training, and the fastest growth is machine-generated data from agents working 24/7. Expect exponential increases in telemetry, logs, and agent-to-agent outputs - and plan your data pipelines accordingly.
Inside Cisco's build: chips, routers, optics
The P200 targets scale-out AI where single-data center limits no longer apply. Think multi-data center clusters stitched together over long distances with deep buffering and new coherency requirements. The 8223 aims to move traffic at that scale reliably.
On the physical layer, Patel called out a hard pivot: copper isn't enough, and classical optics are straining. Coherent optics will matter more as AI clusters sprawl and east-west traffic explodes. If you're planning fabrics for 2026, budget for this shift.
AI is writing the code - your bottleneck moves
"Seventy percent of all AI products now being developed at Cisco are using code that's generated by AI," Patel said. He expects "at least close to a half a dozen products" with 100% AI-written code by 2026. Humans still matter - for specs and code reviews - but the constraint moves from writing code to reading and validating it.
Translate that to your SDLC: tighter specs, stronger review workflows, heavier automated testing, and traceability from requirement to release. Speed will come from how well you review and verify, not how fast you type.
What IT and development teams should do now
- Capacity and topology planning. Model for multi-data center "ultra-clusters" with inter-DC latency and packet buffering in mind. Revisit power and thermal budgets. Evaluate your optical roadmap and coherent optics readiness with your vendors.
- Secure-by-default AI. Make security a deployment gate, not a best effort. Adopt an AI risk framework, require model and dataset provenance, enforce policy-based data access, and red-team agents before they touch production. Consider NIST's AI Risk Management Framework for structure: NIST AI RMF.
- Modernize the SDLC for AI-generated code. Put spec-first workflows in place, add AI-assisted code review, stack static/dynamic analysis, and expand unit, integration, and property-based tests. Track AI contributions for audit and compliance.
- Data strategy for 2026. Stand up synthetic data pipelines with clear quality gates and privacy controls. Prepare storage and governance for machine-generated data growth (retention, labeling, deduplication, and lineage).
- Networking for AI traffic patterns. Plan congestion control and QoS for bursty east-west flows. Validate routing, buffering, and failure domains for long-haul interconnects. Align with hardware that supports scale-out AI workloads.
Why this matters
If Cisco is right, 2026 will reward teams that can ship agentic apps safely, scale across multiple data centers, and keep verification ahead of code generation. The gap won't be who can generate code - it will be who can prove it's correct, secure, and compliant at speed.
Resources
- NIST AI Risk Management Framework: nist.gov
- Build skills for AI-driven development and secure SDLC: AI Certification for Coding
Your membership also unlocks:
