Get tomorrow's brief in your inbox
Today: Cloudflare cut 1,100 jobs despite record revenue, claiming AI made those roles obsolete. Meta and Google are racing to build AI agents that execute tasks on your behalf. OpenAI published its internal playbook for running coding agents safely in production environments.
Cloudflare became the latest company to announce massive layoffs while posting record revenue, cutting 20% of its workforce (1,100 people) and attributing both outcomes directly to AI adoption. The company reported $639.8 million in quarterly revenue, up 34% year over year, while simultaneously claiming these cuts "are not a cost-cutting exercise" but rather "about defining how a world-class company operates in the agentic AI era."
Cloudflare says AI made 1,100 jobs obsolete, even as revenue hit a record high
CEO Matthew Prince said the company was initially cautious about adopting AI internally, but reached a tipping point last November when teams began seeing "massive productivity gains, team members that were two, 10, even 100 times more productive than they had been before." Cloudflare's internal AI usage has increased by more than 600% in the last three months alone. The entire R&D team now uses the company's Workers platform for vibe coding, with 100% of code reviewed by autonomous AI agents before deployment.
For small business owners, this is a crucial data point. Cloudflare is demonstrating what happens when AI tools move from experimentation to production scale. The company is cutting people from all teams except quota-carrying salespeople, suggesting that even complex roles in engineering, HR, finance, and marketing can be automated or dramatically reduced with AI assistance. If a $2.5 billion revenue company can run leaner with AI, the same logic applies to your business. The question is whether you'll capture those productivity gains before your competitors do.
Google Flow for small business ads
Google launched The Small Brief, an initiative using its Flow AI creative studio to help ad industry legends build studio-quality campaigns for local businesses. The goal is to demonstrate how AI can help small businesses create big-brand advertising without the big-brand budget.
See what happens when creative legends use AI to make ads for small businesses
Google will reveal the final campaigns in June. If you run a service business and have been intimidated by video or display advertising because of production costs, this is worth watching. Flow represents Google's bet that generative AI can democratize creative production. No pricing or public access announced yet.
CyberSecQwen-4B: Local cybersecurity AI
A new 4B-parameter model fine-tuned specifically for defensive cybersecurity tasks, designed to run locally on a single consumer GPU (12GB VRAM). CyberSecQwen-4B matches the performance of larger 8B models on cyber threat intelligence tasks while using half the parameters, making it deployable in air-gapped environments where sending security data to hosted APIs is unacceptable.
CyberSecQwen-4B: Why Defensive Cyber Needs Small, Specialized, Locally-Runnable Models
This matters for small businesses handling sensitive data (healthcare, legal, finance) or operating in regulated environments. Hosted AI APIs create compliance risks when you paste customer data, credentials, or vulnerability reports into them. A local model that runs on your hardware keeps sensitive evidence internal while still providing AI-powered threat analysis. The model is Apache 2.0 licensed, meaning you can use it commercially without restriction.
ServiceNow alternatives guide
Zapier published a detailed breakdown of ServiceNow alternatives for businesses that don't need (or can't afford) the enterprise ITSM platform's six-figure price tag and months-long implementation cycle. Options include Zapier itself for AI orchestration, Jira for dev teams, Freshservice for lightweight ITSM, and Atera for IT automation.
The 8 best ServiceNow alternatives in 2026
If you've been quoted ServiceNow and felt sticker shock, this guide breaks down which slice of that feature set you actually need and where to get it cheaper. Most small businesses don't need enterprise ITSM; they need workflow automation, ticketing, or project tracking, and these tools deliver that without the overhead.
OpenAI's Codex safety playbook
OpenAI published the internal controls it uses to run Codex, its AI coding agent, safely in production. The playbook covers sandboxing, approval policies for risky actions, network restrictions, auto-review modes for routine tasks, and agent-native telemetry to understand what Codex did and why.
Running Codex safely at OpenAI
Key takeaways for small businesses deploying AI agents: (1) Define clear technical boundaries for what agents can access. (2) Make low-risk actions frictionless, but require approval for higher-risk operations. (3) Restrict network access to known-good destinations and require approval for unfamiliar domains. (4) Manage authentication through secure credential stores and enterprise workspace controls. (5) Block or require approval for dangerous shell commands while allowing common benign operations.
This is the blueprint for running agents in production without letting them accidentally delete your email, deploy broken code, or expose credentials. If you're experimenting with Claude Code, GitHub Copilot, or other coding agents, these controls are worth implementing now.
Cloudflare's AI adoption timeline
Cloudflare's CEO revealed that the company's internal AI tipping point came in November 2025, when productivity gains became impossible to ignore. Usage spiked 600% in the last three months, driven by developers using AI coding tools and employees across all departments running "thousands of AI agent sessions each day" to complete their work.
Cloudflare says AI made 1,100 jobs obsolete, even as revenue hit a record high
The lesson: AI adoption accelerates once teams see real productivity gains in their daily work. Cloudflare didn't mandate AI use; teams adopted it organically once the tools proved useful. If your team isn't using AI tools yet, the question is whether you're missing productivity gains or whether your processes genuinely don't benefit. Either way, competitors are experimenting, and you need data.
CyberSecQwen-4B: Free, Apache 2.0
The new 4B cybersecurity model is released under Apache 2.0 license, meaning it's free to use, modify, and deploy commercially. It runs on consumer hardware (12GB VRAM), making it accessible to small businesses without GPU clusters.
CyberSecQwen-4B: Why Defensive Cyber Needs Small, Specialized, Locally-Runnable Models
This is a significant cost-reduction opportunity if you're paying per-call for threat intelligence, CVE classification, or security Q&A through hosted APIs. A local model has zero marginal cost after deployment.
Meta and Google building AI agents
Both Meta and Google are reportedly building AI agents designed to carry out everyday tasks for users. Meta is working on a "highly personalised AI assistant" for daily tasks, while Google is developing a "24/7 personal agent for work, school and daily life, powered by Gemini." Both efforts follow OpenClaw's viral success earlier this year.
The Tech Download: Meta, Google enter AI agent race as 'agentic wars' heat up
The driver is clear: agents represent the shift from AI as a cost center to AI as revenue infrastructure. For companies with large advertising and ecommerce businesses, agents that conduct transactions become a major value driver. For small businesses, this means AI assistants will increasingly be able to book appointments, place orders, manage calendars, and complete purchases on your behalf, which changes how you interact with customers and vendors.
Security and governance remain works in progress. A Meta employee went viral in February after OpenClaw deleted a large amount of emails autonomously. The risk management challenge is shifting from AI that says the wrong thing to AI that does the wrong thing.
xAI partnering with Anthropic for compute
xAI is providing compute infrastructure to Anthropic as part of a data center deal. Meanwhile, OpenAI announced it's bringing its models to AWS, renegotiating its exclusive Microsoft partnership to meet enterprise demand for multi-cloud deployment.
Microsoft was worried OpenAI would run off to Amazon and 'shit-talk' Azure
For small businesses, this means AI model access is becoming more flexible. If you've been locked into a single cloud provider because that's where your AI models live, multi-cloud availability reduces that friction.
Akamai lands $1.8 billion AI infrastructure deal
Akamai announced a $1.8 billion, seven-year contract with a "leading frontier model provider" for cloud infrastructure services. The company's cloud infrastructure revenue jumped 40% year over year, and its stock soared 20% on the news.
Akamai stock soars 20% on earnings, $1.8 billion AI infrastructure deal
Akamai operates an AI-powered inference cloud across 4,300 locations in 130 countries, positioning it to run AI applications close to users for faster response times. This matters if you're building customer-facing AI features and latency is a concern; distributed inference infrastructure like Akamai's can deliver sub-100ms response times that centralized clouds can't match.
Sony using AI for game animation
Sony shared details on how it's using AI in PlayStation game development. Studios like Naughty Dog and Santa Monica Studio have adopted an AI-powered animation tool called Mockingbird that completes facial animation work in "a fraction of a second" compared to hours of manual labor. Sony emphasizes that "AI is meant to augment capabilities, not to replace" talent.
PlayStation sees AI as a 'powerful tool' to help make games
The lesson for small businesses: automation tools work best when they eliminate repetitive workflows, not when they replace creative decision-making. Mockingbird processes performance capture data faster, but it doesn't replace the actors or animators who create that data. If you're evaluating AI tools, look for tasks that are repetitive, time-consuming, and well-defined, not tasks that require creativity or judgment.
HTML over Markdown for AI output
Simon Willison highlights a shift from requesting Markdown output from AI models to requesting HTML, especially now that token limits have expanded dramatically. HTML enables AI to include SVG diagrams, interactive widgets, in-page navigation, and rich styling that Markdown can't support.
Using Claude Code: The Unreasonable Effectiveness of HTML
Try this: Next time you ask Claude to explain something complex (a security vulnerability, an API architecture, a code review), request the output in HTML and ask for interactive elements, color-coded findings, or inline annotations. You'll get richer, more navigable explanations.
Evaluate AI tools for automation, not magic
Cloudflare's experience shows that AI productivity gains come from automating repetitive workflows, improving software engineering speed, and accelerating quality assurance and 3D modeling. The work still gets done; it just requires fewer people.
If you're trying to decide where AI fits in your business, start with tasks that are repetitive, time-consuming, and clearly defined. Don't expect AI to handle ambiguous, creative, or judgment-heavy work without human oversight.
The AI agent race is accelerating, with Meta, Google, and Microsoft all betting that agents will shift AI from answering questions to executing tasks. Cloudflare's 20% workforce cut based on AI productivity gains suggests this transition is happening faster than most businesses realize. Security and governance remain weak points, but the momentum is clear: companies that figure out how to deploy agents safely will capture significant productivity advantages.
For small businesses, the window to experiment is closing. Competitors are already running AI agents internally, automating workflows, and capturing productivity gains. The question isn't whether AI will change how work gets done; it's whether you'll be ahead of or behind that curve when it happens.