You’re reading AI at the Top - Our Sunday series on how Fortune 500 enterprises use AI in business.
1. The short version
The company: Cisco booked $17.3 billion in fiscal Q4 2026 revenue, up 18%, and $63.3 billion for the year.
The initiative: An in-house assistant (Circuit), AI agents in support, Codex in engineering, and free security tools (section 3).
The return: 145,000 support cases resolved entirely by AI in FY26 and 75 million+ Circuit prompts in Q4, Cisco says. It has disclosed no dollar return.
2. Cisco built its own assistant instead of banning ChatGPT
When ChatGPT arrived, Cisco’s leaders asked, “Should we actually block it?” chief automation officer Srini Namineni told CIO. They chose “a secure alternative.”
Its internal AI assistant, which Cisco calls Circuit, now has more than 96,000 users as of Q1 2026, 90% employee adoption, and a cost of about $10 per user per month. The project launched in late 2023 to consolidate AI into one platform, and staff have had it since 2024. It began with Azure OpenAI and Google Gemini. Namineni says it now adds a requested model within a couple of weeks. Later came prompt sharing, HR tasks such as booking vacation days, and RAG as a service for querying internal documents.
Circuit runs on Cisco’s Secure AI Factory infrastructure, which “automatically routes each task to the appropriate large language model” to manage token consumption. In August, CEO Chuck Robbins called it “fully embedded in how Cisco operates,” with over 75 million prompts in Q4.
Cisco says Circuit saves engineers about six hours a week and other employees five. In internal polling, 79% said it saves time. Cisco gives no sample size, method or baseline.
🌟 Want to be featured in the next issue? Reach out with your best AI use case and we’ll spotlight it.
3. What Cisco gives away, and how to try it this week
Cisco publishes free tools that catch agents reaching further than they should.
Skill Scanner (2.5k stars when we read it on 26 October 2026) checks downloaded agent skills for prompt injection and malicious code.
Try it: Scan your team’s skills.MCP Scanner (1.1k stars when we read it on 26 October 2026) checks MCP connectors.
Try it: Have IT scan yours.Antares (~320 likes and ~1,500 downloads last month when we read it on 26 October 2026) locates known vulnerabilities in code.
Try it: Request access and run the quickstart.
What to ask for back: Findings by severity, what was scanned, what the scanner cannot catch, and an owner and fix date per finding.
Built by Neatprompts from GitHub, PyPI Stats and Hugging Face.
Enjoying this read? Share the smartest part with your followers.
Click to Share.
4. Support and engineering carry the biggest numbers
Robbins said 145,000 support cases were resolved entirely by AI with zero human intervention in FY26. The work ran on “generative AI and agentic systems across our customer experience organization,” which Robbins said is “dramatically expediting quote turnaround and case resolution times.” He gave no total case volume.
OpenAI’s case study says Codex writes 95%+ of Cisco’s new AI features, saves 1,500+ engineering hours a month and raised defect-fix throughput 10-15x. It is a vendor’s story about its customer.
As in our Visa issue, the numbers cover adoption and velocity with no dollar return.
Built by Neatprompts from Cisco’s Q4 FY26 prepared remarks, CIO and OpenAI’s Cisco case study.
5. The system prompt your team can test this week
Paste in any internal AI claim, and this prompt tells you what's missing before your board asks.
Role: You are a skeptical CFO auditing one internal AI impact claim before it goes to the board. You do not guess, round up or fill gaps.
First two steps:
1. Quote the metric exactly as written, with its unit and time period. If the period is not stated, write NOT DISCLOSED.
2. Name the denominator or baseline (total cases, total users, prior-year figure). If not stated, write NOT DISCLOSED.
The full version adds seven steps, an output table and four stop conditions.
[Click below to open full prompt]Get more such prompts in the Prompting Playbook (free for you)
6. The headcount story runs alongside the AI story
On 13 May 2026, Robbins told staff Cisco would cut fewer than 4,000 jobs, under 5% of the workforce while investing in “our employees’ use of AI across the company.” The 8-K filed that day puts the plan at up to $1 billion in pre-tax charges. Cisco booked $511 million of restructuring and other charges in the fourth quarter and $693 million across fiscal 2026, and the 10-K counts about 82,400 employees as of 25 July 2026.
Robbins cited the “highest revenue, operating margin, and earnings per employee in thirty years,” though Cisco never says AI replaced those roles.
7. What to take to your board
Every company we've tracked, in one place. Our enterprise AI database lists the in-house tool, the use case and the measured impact for each Fortune 500 company we've covered.
8. Tools and Resources
Tools
promptfoo: Red-teams prompts, agents and RAG apps.
Try it: Run the red-team quickstart on your chatbot.NVIDIA garak: NVIDIA calls it “the LLM vulnerability scanner.”
Try it: python -m pip install -U garak, then garak --list_probes.
Resources
OWASP Top 10 for Agentic Applications 2026: Published 9 Dec 2025 with “more than 100 industry experts, researchers, and practitioners.”
Try it: Check one agent against all ten.OWASP Top 10 for LLM Applications 2025: Prompt Injection ranks first (LLM01:2025).
Try it: Ask your vendor which risks it tests.NIST Cyber AI Profile, preliminary draft: released 16 Dec 2025, and more than 6,500 individuals have joined the community of interest, NIST says.
Try it: Ask your CISO which focus areas your plan covers.
Sources
Cisco reporting and filings
Products and open assets
Outside Sources
✨ Was this issue valuable for you?
Stay curious, {{first_name | readers}}.
PS. Missed the last issue? Read how Visa uses AI internally
