- Anthropic's Mythos is powerful evidence against skeptics who pooh-pooh AI's usefulness, says F5's Jimmy White
- Mythos is the first of what will be a new class of AI models specialized for specific tasks, according to the F5 AI VP
- Telcos can expect drastically changed traffic patterns due to AI
It's a tough month to be an AI skeptic.
Anthropic's Mythos software is bigger than a product announcement, said F5 AI VP Jimmy White. The software is a proof-of-concept that AI can actually do important work.
That's significant, because throughout the three-year AI boom, skeptics have been pounding a drum: AI is just a toy. It's a stochastic parrot. It's spicy autocomplete. Its ability to do real work is overhyped.
And the skeptics had evidence. More than 80% of companies reported zero productivity gains from AI in a February 2026 National Bureau of Economic Research study surveying 6,000 executives. A randomized controlled trial by research nonprofit Metr found that experienced developers using AI coding tools actually took 19% longer to complete tasks, though they believed the tools had made them faster. A meta-analysis published in California Management Review found no robust relationship between AI adoption and aggregate productivity growth.
"Everyone thought, okay, lots and lots of investment, really great models getting better. But where's the rubber meeting the road?" White told Fierce.
Mythos answered the skeptics. A specialized AI model demonstrated exceptional ability to detect vulnerabilities in source code — a task with important practical value.
"This is a really good 'white knight' type of purpose," White said. Its not just AI being used to generate memes or greeting cards. It's AI doing important work.
Other practical AI uses emerge
And additional AI uses are coming forward. In its earnings call Monday, Verizon said AI is delivering a 1,280-basis-point increase in customer satisfaction scores year-over-year, with 85% of network problems autonomously resolved, before customers see them. TM Forum said agentic AI is pushing telcos to Level 4 autonomous networks.
Developers report AI has transformed coding and vastly improved productivity. The New York Times interviewed 70 developers and found they are now describing software in natural language and supervising AI as it writes the code; startups reported 20-100x gains working greenfield code bases while Google, with vast amounts of brownfield code, reported a 10% productivity gain. But it's not all upside — junior developer jobs are disappearing.
AI finds special purpose
White foresees a new generation of targeted AI tools emerging. As frontier models improve, specialized capabilities get carved out of them and offered as targeted services — distinct tools for distinct jobs, rather than one general-purpose AI doing everything adequately.
Design is the next example he points to. Anthropic partnered with Canva to offer high-quality software design-as-a-service, and White described the results as genuinely impressive — the kind of output that becomes possible when you pair a powerful model with a massive library of design patterns as training material.
Beyond the big two
Mythos has also accelerated the competitive landscape. What White calls the "big two" — Anthropic and OpenAI — has become a "big five," with Google's Gemini 4, Meta and xAI now competing in frontier model development. Both Anthropic and OpenAI now offer code vulnerability scanning as a packaged service. Open-source models haven't reached that bar yet, White said, but they will.
F5 has been building its own AI security stack. At AppWorld in Las Vegas in March, the company launched an AI security platform that includes AI Red Team for automated offensive testing, AI Guardrails for runtime model protection and AI Remediate — a tool that automates the pipeline from vulnerability discovery to enforced protection, addressing the shortage of specialized AI security experts.
What this means for telcos
For telcos, accelerated AI adoption means new patterns in network traffic. Nearly all AI traffic today is text. But multimodal AI — especially voice and audio — is gaining adoption fast, and carries substantially higher data volumes, White said.
Also, agentic AI doesn't sleep. Human-driven traffic follows geography and time zones. Agent-driven traffic is constant, he said.
"Instead of there being a follow-the-sun data profile, it'll be a flat profile all the time, globally," White said. For telcos, that means the familiar concept of busy hours may lose its utility, and overall volume will rise significantly as human-driven peaks give way to constant agent-driven demand.
While agentic AI traffic may be consistent, AI training causes "elephant flows," or big bursts of traffic, according to a Backblaze report released Tuesday. The cloud storage vendor also reported neocloud and hyperscaler traffic dropping sharply over winter but rebounding in March, which it could not explain.