Loading reflection...
A synthesis of 10 perspectives on AI, machine learning, models release, models benchmarks, trending AI products
AI-Generated Episode
As AI weaves itself deeper into daily life and work, 2025 ended with a sharp vibe check: dazzling new capabilities on the surface, hard questions about money, safety, and jobs underneath.
OpenAI’s biggest consumer move of 2025 was turning ChatGPT from a chatbot into an app platform — effectively a super-app that talks to the rest of your digital life.
Inside ChatGPT, you can now connect services like Spotify, Canva, Coursera, Booking.com, Expedia, Zillow, Target, Uber, DoorDash and more. Type “Spotify” or “Canva” at the start of a prompt, and ChatGPT pulls in those apps, acting as a natural-language control layer on top of them.
In practical terms, that means:
It’s powerful, but it comes with a privacy tradeoff. Connecting these apps means sharing data like listening history, shopping patterns, travel plans, and more with ChatGPT. Users get convenience and personalization; they also inherit the risks of deeper data coupling between platforms.
This ecosystem push isn’t just about user delight. It’s OpenAI’s bid for distribution and retention in a world where raw model quality is no longer enough to stand out.
Early 2025 looked like peak AI mania: OpenAI raised $40 billion at a $300 billion valuation and is reportedly chasing another $100 billion at an $830 billion valuation. Anthropic hit $183 billion. xAI, DeepMind’s rivals, and a swarm of first-time founders all pulled in multi-billion or unicorn-scale rounds, often before shipping meaningful products.
That capital is fueling an infrastructure arms race. OpenAI, Meta, Alphabet, and Oracle are committing hundreds of billions to data centers, chips, and energy projects like the Stargate JV. But cracks are emerging: financing partners backing out of mega-deals, grid bottlenecks, and political pushback on data center expansion.
At the same time, the magic of new model releases dulled. GPT‑5, Google’s Gemini 3, and other frontier models improved benchmarks, but the leaps felt incremental compared to GPT‑4-era shocks. Upstarts like DeepSeek showed they could rival top labs on reasoning benchmarks at a fraction of the spend.
That combination — infrastructure strain, less awe from each release, and skyrocketing costs — shifted attention from “How big is your model?” to “Where’s the business model?”
Meta’s $2 billion acquisition of Manus, a Singapore-based AI agent startup boasting millions of paying subscribers and over $100 million in recurring revenue, underscores this pivot. For investors increasingly uneasy about Meta’s $60 billion-plus AI infrastructure bill, Manus represents something novel: AI that already makes money and can be woven straight into Facebook, Instagram, and WhatsApp.
If 2023–2025 were about experimentation, 2026 looks set to be the year enterprises pick their AI stack — and start cutting the rest.
VCs focused on the enterprise expect AI budgets to rise, but be concentrated. Many large companies are currently running multiple tools against the same use case, from sales automation to internal copilot-style systems. As pilots yield real data, CFOs are likely to:
That favors products with real moats: proprietary data, domain-specific workflows, or safety and governance layers that make AI deployable at scale. It’s a much tougher environment for lookalike tools that can be replicated by a cloud provider or language model vendor with a feature release.
In other words, enterprise AI spend may soar in 2026 — while many AI startups never see their slice of the pie grow.
Beneath the platform wars and funding rounds lies the most consequential question: what happens to workers?
Research suggests the anxiety isn’t unfounded. An MIT study estimated that 11.7% of jobs in the U.S. could already be automated with today’s AI. Employers are quietly cutting entry-level roles, and some layoffs are explicitly tied to AI investments.
Enterprise investors expect 2026 to be a year where budgets visibly migrate from headcount to automation:
The optimistic framing is that AI strips out busywork and elevates people into more creative, higher-value roles. The pessimistic one is that “busywork” is precisely what entry-level and lower-income workers are paid to do — and nothing concrete is replacing it yet at scale.
Taken together, these stories describe an industry coming down from its sugar high. ChatGPT’s app ecosystem shows how quickly AI is becoming an interface for everything from playlists to paychecks. But the surrounding context — infrastructure strain, a crowded vendor field, intensifying trust-and-safety crises, and looming labor shocks — suggests 2026 will be less about new demos and more about proving durable value.
AI isn’t slowing down. It is, however, starting to answer to reality: business models, regulation, infrastructure, and the people whose jobs and lives sit on the other side of the prompt.