Blog

By mid-2026, half of IT and digital marketing roles have dropped formal degree requirements. 85% of employers now use skills assessments, and 76% find them more predictive than resumes. But only 17% of HR leaders feel ready to implement skills-based hiring at scale. This post covers the real tools, the real data, and the cultural friction that keeps most companies stuck between aspiration and execution.

In December 2025, Andrej Karpathy went from writing 80% of his own code to delegating 80% to AI agents. Stanford payroll data shows employment for developers aged 22-25 has dropped nearly 20% since ChatGPT launched. Entry-level tech hiring fell 25% at major firms. The junior developer pipeline is not just shrinking; it is structurally breaking. This post covers the real numbers, why the ‘apprenticeship’ model of software engineering is collapsing, and what skills still get junior developers hired.

The traditional B2B sales funnel has 7 stages and takes 3-6 months to close a deal. Agentic AI compresses that into 3 autonomous phases and cuts cycle times by 28%. Here is how autonomous revenue engines actually work, who is building them, and where they still break.

ByteDance’s UI-TARS Desktop is an open-source GUI agent that sees your screen and performs actions, no DOM parsing or APIs required. Built on the Qwen2-VL vision-language model, the 72B variant scores 24.6% on OSWorld (beating Claude Computer Use at 22.0%), while the 7B model runs locally on a single GPU. This post covers the architecture, real benchmarks, how it compares to browser-based agents, and what the open-weight approach means for enterprise teams that need data sovereignty.

Corporate legal departments adopted AI at more than double the rate of law firms in 2025, jumping from 23% to 52% active use. 64% of in-house teams now expect to depend less on outside counsel, and 78% see drafting as an insourcing opportunity. The billable hour model creates a structural disincentive for law firms to adopt AI aggressively, while every dollar corporate legal saves on outside counsel drops straight to the bottom line. With the EU AI Act taking full effect in August 2026 and the Colorado AI Act following in June, regulation is catching up to the technology.

IBM announced it will triple US entry-level hiring in 2026, even as AI agents automate the tasks those roles used to perform. The company’s CHRO says the old entry-level job is dead, but the new one, redesigned around AI collaboration, client work, and cross-functional judgment, is more valuable than ever. This is the clearest signal of a two-speed workforce: companies that invest in AI-native juniors now will dominate in five years. Companies that hollow out their pipeline will face a leadership crisis.

The KYA (Know Your Agent) market went from zero vendors to six in under 90 days. Sumsub shipped the first commercial agent-to-human binding product in January 2026. Vouched raised $17M and launched Agent Checkpoint in February. Beltic built on W3C DIDs. Dock.io extended Anthropic’s MCP with identity primitives. This post compares these vendors across verification approach, regulatory coverage, integration complexity, and use cases.

ElevenLabs became the first company to go live with an AI agent insurance policy backed by AIUC-1 certification in February 2026. The certification puts AI systems through 5,835 adversarial tests across 14 risk categories, from prompt injection to data leakage. For enterprises running 3M+ voice agents across customer support, sales, and scheduling, this creates a new risk transfer mechanism: insurable AI agents that shift liability from the deploying company to an underwriter. This post covers what the certification tests, what the insurance covers, who else is getting certified, and what it means for the EU AI Act compliance timeline.

The threat model for digital commerce just shifted. Instead of stolen credit cards, attackers will target stolen or manipulated AI agents. The industry response is a new trust layer: Know Your Agent frameworks, Digital Agent Passports, and tokenized authentication from Visa, Mastercard, and Cloudflare. But a critical liability gap remains.

The UK’s Information Commissioner’s Office published Tech Futures: Agentic AI in March 2025, making it the first data protection regulator anywhere to formally assess the risks of autonomous AI agents. The report identifies eight specific risk areas, from cascading hallucinations to controller/processor role confusion in multi-agent systems. Here is what the ICO found, how it compares to EU approaches, and what it means for companies deploying AI agents in Europe.