AI adoption surged to 84% of developers, and 41% of all code is already AI-generated or AI-assisted, fundamentally changing how software is written and shipped. Yet confidence moved in the opposite direction. Only 29% of developers trust AI-generated code (vs 40% in 2024), and nearly half say fixing “vibe-coded” output takes longer than writing it themselves. The most unsettling insight came from controlled studies: developers using AI completed tasks 19% slower while believing they were 20% faster, exposing a widening gap between perceived productivity and reality.

Cursor redefined the IDE for an AI-first world, forcing every serious developer to follow its lead. GitHub completed its shift from code host to full developer platform, overtaking Jira by bundling collaboration, Copilot, and Spark into a single workflow developers actually want to use. Vercel solidified its role as the default deployment layer for modern web teams, with Next.js dominance and v0 pushing rapid prototyping into the mainstream. Nebius showed hyperscalers aren’t invincible, winning developers by prioritizing simplicity and AI-first infrastructure over bloated service catalogs. Warp modernized the terminal itself, proving even the most entrenched workflows can change when AI is designed in from the start.

Reflection AI signaled a renewed push toward open frontier models, with its meteoric rise framed as a direct response to DeepSeek and growing concern over closed AI dominance. Graphite and Greptile emerged as essential counterweights to AI code generation, turning review and quality control into scalable, AI-native systems. Browserbase quietly became foundational as agents moved from demos to production, providing the missing infrastructure for machine-driven web interaction. Mastra captured the familiar open-source arc early, rapidly building community and enterprise traction by giving developers real control over how agentic systems are built.

As hype cooled, the industry openly questioned whether AI valuations and infrastructure costs were justified by real revenue, with even Sam Altman acknowledging bubble risks. Microsoft quietly emerged as the biggest power broker by partnering with Open AI, and later Anthropic and Nvidia, positioning itself as the default AI infrastructure layer. DeepSeek and Jensen Huang’s remarks shattered assumptions of US dominance, proving China is much closer in frontier capability and moving faster in adoption. OpenAI, long seen as untouchable, showed visible cracks and declared code RED as competition intensified and burn rates surged, prompting internal emergency signals. And with the $500B Stargate project, AI officially crossed into national infrastructure, turning model development into a geopolitical and economic bet with enormous stakes.

Perplexity reshaped browsing into an answer-first experience, changing expectations for how information should be found and consumed. Browserbase pushed browsing behind the scenes, enabling agents to interact with the web without human-facing interfaces. The Browser Company reimagined the browser as a workspace, integrating AI into everyday interactions rather than bolting it on. OpenAI’s Atlas treated the browser as an AI-native operating layer, centered on memory, agents, and task completion. Meanwhile, Dia proved that intelligent browsing can exist without surveillance, highlighting a growing shift toward privacy-first design.

AI trained on the internet produced pure brainrot, from viral Tung Tung Sahur TikTok memes and gorilla vlogs, to Queen Elizabeth wresting in WWE. Meanwhile, one engineer in India pushed hustle culture to parody by secretly working in 4 YC startups at once (and was only discovered through a viral post on X). LLMs meant to transform productivity ended up faking spoiled Uber Eats orders and damaged toys for refund scams. Startups leaned fully into rage bait, turning “Stop Hiring Humans” into a profitable marketing strategy. And to top it off, two AI billionaires publicly feuded like teenagers on X, reminding everyone that even world-changing technology can’t outgrow human pettiness.

A vibe-coded app designed as a safe space placed #1 on the App Store, only to leak IDs and private messages at massive scale. Grok showed how quickly an AI embedded in a social platform can spiral into harmful ideology. You’d think companies have learned their lesson with AI and Twitter since that Microsoft incident. Meanwhile, McDonalds made efforts to implement AI in their hiring, only to expose data of millions of their job applicants. In Hong Kong, deepfakes made even live video calls untrustworthy, with one bank employee sending $25 million dollars in minutes without realizing they were talking with a fake ‘CFO’. Looming over everything was the rise of AI-powered cyberattacks, where scams became automated, hyper-personalized, and more effective than human-driven fraud.

2025 showed us what happens when AI moves faster than our instincts, processes, and safeguards. Adoption exploded, tools reset expectations, and new infrastructure quietly became essential, while trust, security, and certainty lagged behind. We laughed at the absurdity, worried about the consequences, and watched the internet, development, and browsing all get redefined in real time. The takeaway is simple: next phase of AI won’t be about building more, but about building better. We’ll be taking a deep dive into what the future holds for AI in our next week’s issue, so stay tuned!

Till next time,