PLUS: Trump halts AI order and a new agent-powered Kanban board
Happy reading
The honeymoon phase of cheap, subsidized AI appears to be over. As companies like Microsoft and Uber deploy the technology at scale, they're discovering that runaway costs are challenging the popular narrative of AI as a simple and cheap replacement for human labor.
The explosion in token consumption, largely driven by the rise of powerful AI agents, is the primary culprit. As providers shift away from unsustainable flat-rate plans, how will this new economic reality shape the future of AI adoption and force a more critical focus on return on investment?
In today's Next in AI:
The rising costs of enterprise AI
Trump pauses key AI executive order
An agent-powered Kanban for developers
AI recreates pilot voices from an image
AI's Real Cost Problem
Next in AI: The era of subsidized AI appears to be ending as major companies like Microsoft and Uber discover that using the tech at scale is blowing up budgets, challenging the narrative that AI is a cheap replacement for human labor.
Explained:
Enterprise teams are hitting a cost wall with token-based pricing, with reports showing Uber exhausted its entire 2026 AI coding budget in just four months after encouraging widespread adoption.
This is driven by the rise of powerful, multi-step AI agents, which can consume vastly more tokens per task and create a need for new governance frameworks to manage costs and complexity.
The cost paradox is simple: even as individual token prices fall, total consumption is exploding, leading to inefficient AI token usage and making flat-rate AI features financially unsustainable for providers.
Why It Matters: This reality check is forcing a crucial shift from "add AI everywhere" to focusing on use cases with a clear return on investment. Expect to see fewer flat-rate plans and more metered, credit-based pricing models as the true cost of AI is passed on to users.
The AI Order That Wasn't
Next in AI: President Trump hit pause on signing a landmark AI executive order this week after a last-minute warning from insiders that it could stifle U.S. innovation. The order was set to create a voluntary government review process for new, powerful AI models before their public release.
Explained:
The draft of the order outlined a voluntary review system, allowing developers to submit advanced models to federal agencies up to 90 days before launch.
Former AI czar David Sacks reportedly convinced Trump to delay the signing, arguing the reviews could slow development and harm America's lead in the AI race with China.
Industry reception was mixed, with some companies like OpenAI supporting the idea while others pushed to shorten the review window from 90 days to just 14.
Why It Matters: This eleventh-hour reversal puts the intense debate between rapid innovation and AI safety squarely on the president's desk. The episode highlights the challenge of crafting policy that secures advanced AI without ceding the competitive edge to global rivals.
The Agent-Powered Kanban
Next in AI: A new open-source tool called KanBots brings AI agents directly into the developer workflow. It uses a Kanban board to dispatch and manage multiple agents that autonomously handle coding tasks in parallel.
Explained:
Parallel agents work on different tasks simultaneously, each operating in an isolated git worktree to avoid conflicts and allow for safe experimentation.
It embraces the Kanban metaphor for AI, featuring an "autopilot" mode where personas—like product, engineer, and QA—collaborate to split work and review output.
The tool is local-first to keep your code on your machine and includes live cost tracking to prevent surprise bills and a pre-push hook for safety.
Why It Matters: This shifts AI-driven development from a simple chat interface to a structured project management system. It provides a scalable and predictable way to integrate teams of AI agents into real-world software cycles.
From Spectrogram to Speech
Next in AI: In a startling display of AI's power, internet users recreated pilot voices from a fatal plane crash using only a public spectrogram image. The incident prompted the National Transportation Safety Board (NTSB) to suspend public access to its entire accident database.
Explained:
A spectrogram is a visual representation of sound, and individuals used AI models and modern algorithms to convert the image back into audio, with one person claiming it took just 10 minutes.
The reconstruction circumvents a federal law enacted in 1990 that prohibits the NTSB from publicly releasing cockpit audio recordings to protect the privacy of air crews.
The NTSB's swift action to shut down its public docket system highlights a new challenge for government agencies: data previously considered non-sensitive can now be transformed by AI to reveal restricted information.
Why It Matters: This event demonstrates that AI can extract sensitive information from data formats previously thought to be safe for public release. It forces a critical re-evaluation of public data policies in an era where AI can uncover what was intended to remain hidden.
AI Pulse
Samsung agreed to a landmark deal with its union that will pay out an average bonus of $340,000 to chip division employees, funded by soaring AI-driven profits.
California issued an executive order directing state agencies to study AI's impact on the workforce and create new job-training programs in response to ongoing tech layoffs.
GitHub introduced new policies and features to combat a wave of low-quality, AI-generated code contributions, a phenomenon some engineers are calling “vibe slop.”
Google's new AI-powered Search returned a completely blank summary for the query "disregard," highlighting early, public-facing issues with its aggressive rollout of AI Overviews.
Elon Musk proposed that a "Universal High Income" funded by the government is the best way to address widespread unemployment caused by AI.