AI News Archive: June 1, 2026 — Part 9
Sourced from 500+ daily AI sources, scored by relevance.
- Google’s chief economist backs Ottawa’s ‘AI for all’ strategy
AI Minister Evan Solomon expected to launch a national policy on the technology this week
- How Edge Intelligence Can Be Used In Safety-Critical Environments
Edge intelligence can power physical AI systems, enabling real-time perception and action in the physical world.
- Anthropic selects Birmingham for national small business tour
The tour consists of free, half-day AI training events in just 10 cities across the country. Birmingham is one of those cities.
- Will AI Help or Overwhelm Students? Teachers Weigh In
Even as teachers across the country experiment with AI, many are skeptical of its role in classrooms, and whether it will undermine student learning.
Score: 20🌐 MovesJun 1, 2026https://www.edweek.org/technology/will-ai-help-or-overwhelm-students-teachers-weigh-in/2026/06 - Why Pope Leo and Tony Blair's treatises on AI have captured global attention
Why Pope Leo and Tony Blair's treatises on AI have captured global attention The National
- UOB and FPT sign MoU to advance AI, technology transformation and financial innovation opportunities
UOB and FPT sign MoU to advance AI, technology transformation and financial innovation opportunities The Straits Times
- AI Agent Culture: What Does it Mean For Your Lean Team?
Supercharge your entire workflow by welcoming digital assistants to your team to do the heavy lifting for you.
- Market Logic Network Builds AI Search Visibility Systems for Modern Businesses
Market Logic Network Builds AI Search Visibility Systems for Modern Businesses azcentral.com and The Arizona Republic
- Import AI 459: AI oversight is difficult; scaling laws for protein folding models; and pricing the extinction risk of AI systems
Do you feel as though you are living in a revolution?
- Google’s next AI may not be good for you
Google is increasingly using AI to package information for users, raising questions about transparency, accuracy, and trust.
- Digital Science launches AI-Assisted Profile Curation in Symplectic Elements
Digital Science launches AI-Assisted Profile Curation in Symplectic Elements EurekAlert!
- AI could democratize access to expert coaching, sports tech entrepreneurs say
Industry leaders see AI as a “complementary piece” of the puzzle alongside actual coaches. The post AI could democratize access to expert coaching, sports tech entrepreneurs say first appeared on BetaKit .
Score: 19🌐 MovesJun 1, 2026https://betakit.com/ai-could-democratize-access-to-expert-coaching-sports-tech-entrepreneurs-say/ - Boston Celtics alum Jeff Teague has some thoughts about the NBA using AI to officiate
Boston Celtics alum Jeff Teague has some thoughts about the NBA using AI to officiate Celtics Wire
- OmAI Launches OttoBox AI Video Creation Assistant at BEYOND Expo 2026
OmAI unveiled OttoBox, an AI-native video creation assistant powered by its in-house OmModel multimodal model, reducing rough cut editing from 8 hours to 30 minutes.
Score: 19🌐 MovesJun 1, 2026https://pandaily.com/om-ai-launches-otto-box-ai-video-creation-assistant-at-beyond-expo-2026 - Generative AI for software engineers: How to build the right AI stack
Learn how Glean helps build a generative AI stack for software engineers with shared context, guardrails, and workflows
- Reinforcement learning is an infrastructure problem
What we've seen helping teams run Reinforcement Learning at scale on Modal. Plus an open-source library to skip the scaffolding.
- I Built a RAG System That Never Hallucinate — Here’s the Exact Architecture
The problem every engineering team has but nobody talks about You’re on-call. It’s 2am. The checkout service is throwing 503s. You need to know: which service is actually broken, who owns it, what changed in the last 48 hours, and whether there’s a related open ticket — all in under 60 seconds. The answer exists. It’s scattered across Jira, GitHub, Confluence, and your log aggregator. None of those tools talk to each other, and none of them understand your question. This is the problem I set out to solve with Aksibu Mini : a fully open-source engineering context engine that aggregates fragmented engineering data into a unified knowledge base, answers operational questions in natural language, and — most importantly — never makes things up. Two systems in one Aksibu Mini has two distinct pipelines that I’ll walk through in depth. Pipeline 1: GraphRAG — knows your engineering stack. Jira tickets, GitHub commits, Confluence docs, application logs. It builds a knowledge graph, understands who owns what, how services depend on each other, and can trace an incident from a log line back to the commit that introduced it. Pipeline 2: Cited Document Search — knows your documents. Point it at a folder of a million PDFs and ask anything. Every single sentence in the answer comes with a `[1]`-style citation pointing to the exact paragraph it came from. If something isn’t in your documents, it says so instead of guessing. Both pipelines run at near-zero cost. The entire GraphRAG index on 18 records costs ~$0.01 to build. The document search pipeline runs entirely on local models — zero API cost for retrieval. Pipeline 1: GraphRAG on a budget The first challenge was cost. A naive implementation would send every engineering record to a frontier LLM for entity extraction — at real scale that’s thousands of dollars. Instead I built a tiered extraction system. The three tiers: Tier 1 — Regex rules (free). A rule-based extractor handles roughly 70% of records. It pattern-matches service names, ticket IDs, engineer names, and common relationship phrases. Zero API calls, zero cost. Tier 2 — Cheap LLM. Only records with unresolved text get escalated. A fast, cheap model (Gemma 27B) handles these for ~$0.0002 per record. Tier 3 — Frontier LLM. Only complex records trigger this: Confluence docs (rich prose, many implicit relationships), records over 800 characters with fewer than 2 entities found, or records with entities but zero detected relationships. Cost ~$0.003 per record. The result: first build costs ~$0.01. Every subsequent run with no changes costs exactly $0.00 because everything is cached in SQLite. The knowledge graph Entities extracted from all four sources feed into a `networkx.DiGraph`. Nodes represent services, engineers, tickets, commits, log lines, and docs. Edges represent relationships: `owns`, `affects`, `fixes`, `depends_on`, `caused_by`, `modified`. checkout-api ── owns ──→ Rahul Sharma checkout-api ── affects ──→ JR-441 JR-441 ── caused_by ──→ sha:a3f8c2 sha:a3f8c2 ── modified ──→ checkout-api When you ask “Why is checkout broken?”, the agent doesn’t just do a keyword search. It traverses the graph: starts from the service node, BFS-expands 2 hops, finds related incidents, recent commits, and the on-call owner — all in one query. Agentic retrieval with a cost cap The retrieval layer uses a tool-use loop. The LLM decides which tools to call based on the question, and has six available: `semantic_search`, `get_service_context`, `find_owner`, `get_recent_changes`, `get_active_incidents`, `get_dependencies`. Simple questions (under 8 words, starting with “what is” or “define”) skip the agent entirely and go straight to vector search — no tool calls, minimum cost. Complex questions run through the loop, capped at 7 tool calls. Hit the cap and the system forces a synthesis with whatever it has. No runaway API cost. Pipeline 2: Cited Document Search at scale This was the harder problem. The GraphRAG pipeline works on a fixed, structured dataset. The document search pipeline needs to work on anything — research papers, architecture docs, legal contracts, a folder of a million PDFs — and return answers where every claim is provably grounded. Why single-stage retrieval fails at scale: Most RAG tutorials show the same approach: embed your documents, store them in a vector database, retrieve top-k, send to LLM. This breaks down in two ways: Semantic gap. Vector search finds semantically similar text. But if you ask “What does CONF-1023 say about the Redis failover?” and your doc literally contains the string “CONF-1023”, your vector embedding likely won’t rank it first — exact token matches are drowned out by semantic noise. Precision gap. At 1 million documents, top-5 vector results are noisy. The LLM will receive irrelevant context and hallucinate connections between unrelated information. The three-stage hybrid retrieval Stage 1: BM25 + vector in parallel. BM25 (a classical TF-IDF variant) excels at exact term matching. Dense vector search excels at semantic similarity. Run both, get 50 candidates each. Stage 2: Reciprocal Rank Fusion. Merge the two lists without requiring score normalisation. The formula is simple and elegant: A chunk appearing in both lists at rank 3 and rank 7 scores higher than a chunk appearing in only one list at rank 1. This rewards cross-list agreement — which is exactly the signal you want. Stage 3: Cross-encoder reranking. The bi-encoder that built your vector index processes query and document independently. A cross-encoder processes them *together* , enabling full attention between query tokens and document tokens. This is far more accurate but too slow to run on all chunks — so I run it only on the top-20 RRF candidates, producing top-5 final chunks in milliseconds. The model is 22MB, runs locally, costs nothing, and was trained on 8.8 million query-passage pairs from Microsoft’s MS MARCO dataset. Citation enforcement: the key to zero hallucination Retrieval quality gets you 80% of the way there. The last 20% is prompt architecture. The system prompt is a hard constraint: You are a precise document analyst. Every factual claim MUST be followed immediately by [N] where N is the source number. You MUST NOT state anything that cannot be attributed to the provided sources. If the answer is not in the sources, say: “The provided documents do not contain enough information to answer this question.” Each of the 5 retrieved chunks is passed as a numbered source with its filename and position: [1] (from: architecture.pdf, chunk 3 of 12) Redis is used for session caching. All user sessions and cart contents are stored with a 30-minute TTL… [2] (from: incident-runbook.md, chunk 1 of 4) In the event of Redis unavailability, the checkout service falls back to in-memory session storage… The LLM produces: Redis handles session caching with a 30-minute TTL [1]. If Redis goes down, checkout falls back to in-memory storage [2]. Then post-processing validates every `[N]` reference against the supplied chunks. Any citation pointing to a number outside `[1, 5]` — something the LLM invented — is stripped from the response. This is a structural guarantee, not a soft instruction. The LLM literally cannot produce an uncited or fictionally-cited claim that survives post-processing. Scaling to 1 million documents The folder ingestion path is memory-safe by design. Files are processed one at a time from disk — `chunk_file_from_path(path)` reads, chunks, and discards each file before loading the next. Chunks are upserted to ChromaDB in batches of 200. The BM25 index is rebuilt exactly once at the end. ChromaDB’s HNSW index gives O(log n) query time at any scale. A query over 1 million chunks completes in ~5–10ms. The BM25 index holds ~2–4GB in RAM at 500K chunks — above that, the right swap is ElasticSearch or a sparse vector store like SPLADE. What I’d build next? Streaming answers. The LLM currently returns the full response before the UI renders. Adding `stream=True` to the OpenAI client call would let citations appear word by word. Multi-hop citation chains. Right now citations point to chunks. A more powerful version would trace: *”claim → chunk → source document → original data source”* — a full provenance chain from answer back to the raw Jira ticket or commit. ElasticSearch for BM25 at extreme scale. The in-process `rank_bm25` pickle is convenient but caps out around 500K chunks. Swapping it for an ElasticSearch instance keeps the same RRF architecture and removes the RAM ceiling. Scheduled re-indexing. The incremental manifest system already tracks which records have changed. Wrapping `python main.py` in a cron job would keep the knowledge graph current without manual intervention. Try it yourself The full source is on GitHub: [github.com/HetPatel1978/aksibu_mini]( https://github.com/HetPatel1978/aksibu_mini ) Setup takes about 5 minutes: git clone https://github.com/HetPatel1978/aksibu_mini Upload a few PDFs on the Document Search tab and ask anything. The citations are live — every `[1]` is clickable and expands to show the exact paragraph the answer came from. *If you found this useful, the repo is open-source and PRs are welcome. The architecture is deliberately minimal — the goal was to show what’s possible without any paid vector database, managed embedding API, or LLM framework.* I Built a RAG System That Never Hallucinate — Here’s the Exact Architecture was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- Tesla’s Irish revenues and profits down last year as car sales fall
Elon Musk’s carmaker sees bright start to 2026 as EV sales rise across the board
- No One Owns AI Spend: Three Preconditions For Distributed Accountability
Decisions are already distributed. Accountability isn't. That's the gap to close.
- Veeam’s Securiti AI named a leader and fast mover in GigaOm’s 2026 DSPM Radar with top scores among evaluated vendors
GigaOm’s 2026 DSPM report recognized Securiti AI as a Leader for the third consecutive year, supported by Data Command Graph visibility, automated remediation, and broad hybrid-environment coverage
- Allegedly trashing Airbnbs to test robots puts startup in legal trouble
Lawsuit seeks $12,000 from startup that allegedly damaged home in robot tests.
Score: 18🌐 MovesJun 1, 2026https://arstechnica.com/ai/2026/06/allegedly-trashing-airbnbs-to-test-robots-puts-startup-in-legal-trouble/ - Metacognitive laziness and AI: researchers propose a scale to measure a growing educational concern
Metacognitive laziness and AI: researchers propose a scale to measure a growing educational concern EurekAlert!
- AI-Driven car-buying process demands more accurate data
AI-Driven car-buying process demands more accurate data Automotive News
Score: 18🌐 MovesJun 1, 2026https://www.autonews.com/resource_center/Experian/an-ai-driven-car-buying-process-demands-accurate-data/ - I Build Robots That Restore Ecosystems. Here’s Why AI Is Being Used for the Opposite
As a founding team member of Google X with deep expertise in autonomous systems, I’m here to tell you why we’re choosing the wrong path for robotics—and what we can to to change course.
- When condo boards attack + Toronto’s rich are turning to AI to fight home invasions
When condo boards attack + Toronto’s rich are turning to AI to fight home invasions Toronto Star
- How an AI-powered fintech startup built a website AI actually trusts
Why Chargeflow invested in a website that's both visually distinctive and optimized for AI search
Score: 18🌐 MovesJun 1, 2026https://webflowmarketingmain.com/blog/how-an-ai-powered-fintech-startup-built-a-website-ai-actually-trusts - AI innovation moves fast. Security must help it move faster.
Organizations are using copilots, autonomous agents, and AI-driven workflows to move faster, make smarter decisions, improve productivity, and unlock new ways of working. In many industries, the winners will not simply be the companies that adopt AI, but the ones that can operationalize it quickly, confidently, and at scale. But accelerated innovation also introduces a new kind of risk. Innovation Is Accelerating. So Is Complexity. AI agents are not passive tools. They can reason, act, access systems, invoke applications, interact with sensitive data, and execute workflows. In effect, they are becoming a new class of non-human identity inside the enterprise. And unlike traditional users, these agents often operate dynamically, across SaaS applications, browsers, endpoints, APIs, and cloud environments — sometimes outside the view of IT and security teams. That creates a critical challenge: how can organizations embrace AI innovation without creating unmanaged risk? The answer is not to slow innovation down. Security should not be a detractor from AI adoption. It should be the foundation that allows organizations to move faster with confidence. The Foundation of Trusted AI Starts with Identity To do that, enterprises need a modern approach to agentic security — one built around visibility, governance, and protection. First, organizations need to know which AI agents exist, who owns them, what they can access, and how they connect to sensitive data. Without that context, teams are innovating in the dark. Second, organizations need governance that can keep pace with agentic activity. AI agents should be treated as first-class identities, with clear human ownership, lifecycle management, least-privilege access, and audit-ready controls. This gives security and compliance teams the oversight they need, while giving business and innovation teams the guardrails they need to move forward. Finally, organizations need protection and response capabilities that operate at machine speed. If an agent behaves unexpectedly, accesses the wrong resource, or becomes compromised, manual response may be too slow. Real-time monitoring, risk scoring, and automated remediation help contain threats before they become larger incidents. Enabling the Business to Create Safely This is where the SailPoint Agentic Fabric becomes a meaningful enabler. By helping organizations discover, govern, and protect AI agents across the enterprise, it supports a more confident path to AI adoption. It brings identity context to the center of agentic security, helping organizations understand not only what an agent is doing, but who owns it, what it can access, and how risk should be managed. The real opportunity is not to choose between speed and safety. It is to make security the reason AI innovation can scale. In the AI era, competitive edge will belong to organizations that can create boldly, move quickly, and protect intelligently. Agentic security is what makes that possible — helping businesses turn AI from a source of uncertainty into a trusted engine for innovation. If you’re ready to see how these principles work in practice, our latest white paper offers a comprehensive guide to making agentic security a reality.
Score: 18🌐 MovesJun 1, 2026https://www.cio.com/article/4179462/ai-innovation-moves-fast-security-must-help-it-move-faster.html - RAG Is Not Machine Learning, and the ML Toolkit Solves the Wrong Problem
Enterprise Document Intelligence [Vol.1 #3] - Why the ML toolkit (hyperparameter sweeps, train/test splits, explainability frameworks) solves the wrong problem, and what to use instead The post RAG Is Not Machine Learning, and the ML Toolkit Solves the Wrong Problem appeared first on Towards Data Science .
Score: 18🌐 MovesJun 1, 2026https://towardsdatascience.com/rag-is-not-machine-learning-and-the-ml-toolkit-solves-the-wrong-problem/ - REDnote Launches AI Search Assistant 'Diandian' on PC
RED (Xiaohongshu) officially launched its AI search assistant "Diandian" on PC on May 29, accessible directly via a dedicated domain. The AI assistant transform...
Score: 18🌐 MovesJun 1, 2026https://pandaily.com/re-dnote-launches-ai-search-assistant-diandian-on-pc - Can India reap AI gains?
That hinges on having ownership of AI capability
Score: 18🌐 MovesJun 1, 2026https://www.thehindubusinessline.com/opinion/can-india-reap-ai-gains/article71049724.ece - LVX Bets On AI To Transform Private Market Dealmaking
Bengaluru-based VC firm LVX (formerly LetsVenture) has unveiled Elvix, an AI-powered intelligence platform designed to simplify deal evaluation and decision-making…
Score: 18🌐 MovesJun 1, 2026https://inc42.com/buzz/lvx-bets-on-ai-to-transform-private-market-dealmaking/ - Return Helper raises $4M Series A to turn cross-border ecommerce returns into profit with AI [Sponsored]
Return Helper, a tech-first provider of cross-border ecommerce returns solutions, has closed a $4 million Series A funding.The round brings together a mix of strategic and financial backers with deep ...
- ChinAI #361: What can CANN do for China's independent compute capacity?
Greetings from a world where…
- Untapped Potential: Why using AI to reach Spanish-speaking buyers is so important
Untapped Potential: Why using AI to reach Spanish-speaking buyers is so important Automotive News
- BIBF trains more than 1,600 Bahrainis to boost workplace productivity through AI
The milestone reflects growing demand for practical skills that help professionals work more efficiently, improve decision-making, and create greater value within their organisations
- Estate agents under fire for using AI pictures in property listing
Estate agents under fire for using AI pictures in property listing The Telegraph
Score: 16🌐 MovesJun 1, 2026https://www.telegraph.co.uk/business/2026/06/01/estate-agents-under-fire-ai-pictures-property-listing/ - The 5 Affordability Segments and How AI Can Help - Automotive News
The 5 Affordability Segments and How AI Can Help - Automotive News Automotive News
Score: 16🌐 MovesJun 1, 2026https://www.autonews.com/sponsored/webinars/5-affordability-segments-and-how-ai-can-help/ - Unlocking Revenue in an AI Age
Unlocking Revenue in an AI Age Fortune
Score: 16🌐 MovesJun 1, 2026https://fortune.com/videos/watch/Unlocking-Revenue-in-an-AI-Age/05e22e8d-92ae-467c-b24a-1ae72fa5663e - We have to get the industry out of proof of concept fatigue in deploying AI: Sandeep Dutta, President, AWS India and South Asia
In his first interview since taking over as AWS President for India and South Asia, Sandeep Duta spoke to businessline about his key focus areas, why he is bullish, and plans to build ‘from India for India and the world’
- Strique unveils self-learning AI platform that optimises D2C growth, revenue and ROAS at Meta day
Strique has launched an autonomous performance marketing execution platform for D2C brands, addressing the gap between knowing what to change and implementing it. This AI-powered system connects to various platforms to audit, reallocate budgets, generate creatives, and execute campaigns, aiming to continuously optimize revenue and performance by learning from every marketing cycle and customer interaction.
- Virtual farms created from real-world tomato farm images
Virtual farms created from real-world tomato farm images EurekAlert!
- This AI Kidnapping Scam Is Every Parent's Worst Nightmare
This AI Kidnapping Scam Is Every Parent's Worst Nightmare PCMag
Score: 16🌐 MovesJun 1, 2026https://www.pcmag.com/explainers/this-ai-kidnapping-scam-is-every-parents-worst-nightmare - Cognizant creates new job roles for next-gen AI talent
Both roles are developed and certified through SkillSpring, Cognizant’s proprietary training platform
- Lantronix Launches SLC 9000 to Secure AI Data Centers and Mission-Critical Networks at Scale
Lantronix Launches SLC 9000 to Secure AI Data Centers and Mission-Critical Networks at Scale markets.businessinsider.com
- BeyondTrust Expands Identity Security Risk Assessment with New Five-Pillar Framework for Human, Non-Human, and AI Identities
BeyondTrust Expands Identity Security Risk Assessment with New Five-Pillar Framework for Human, Non-Human, and AI Identities markets.businessinsider.com
- Berkshire Hathaway deal, startups grapple with AI, the cost of the Iran war and more in Morning Squawk
Here are five key things investors need to know to start the trading day.
Score: 15🌐 MovesJun 1, 2026https://www.cnbc.com/2026/06/01/5-things-to-know-before-the-market-opens.html - I Played MSI's New Claw 8 EX AI+ at Computex. It Won't Be Cheap, But It Just May Be the Gaming Handheld to Beat
I Played MSI's New Claw 8 EX AI+ at Computex. It Won't Be Cheap, But It Just May Be the Gaming Handheld to Beat PCMag
Score: 15🌐 MovesJun 1, 2026https://www.pcmag.com/news/i-played-msis-new-claw8-ex-ai-plus-at-computex-2026 - 4 recs for CIOs to stay on the human side of AI transformation
It’s been recently reported that up to 27 million corporate roles across the Global 2000 are meaningfully exposed to AI-driven elimination, displacement, or fundamental redesign over the next three years. According to the report, however, most organizations sitting on top of these exposures have no coherent plan for what they’re doing with AI, let alone what happens to the people in its crosshairs. While few would argue that bringing AI into the enterprise is the right move, blindly following the pack and eliminating jobs to look good to Wall Street can have disastrous results. A better move is to be bold, fast and responsible, an approach that’s guided organizations like KPMG in their own AI transformations. To push back against the herd mentality and set up your organization for success in the long-term, here are four recommendations for CIOs to navigate the shifting AI and human workforce landscape. Follow your strategy, not the market Over the past decade or more, firms often used the excuse of being in a multi-year digital transformation journey to explain missed earnings and declining revenues. Today, AI transformation is being utilized to similar effect with a surge of 6,550% year-over-year mentions of AI agents in SEC filings. Here’s why cutting jobs without a plan can backfire. By my calculation, out of the 27 million roles at risk, roughly 14% may be permanently lost, amounting to 3.78 million. A further 74% will be reskilled and upskilled, and 12% will be rehired as firms are forced to modify versions of the same roles to repair broken workflows. This means that the bulk of the work over the next three years, or about 86% of this total shift, will be in reskilling, upskilling and rehiring. Rather than job replacement due to AI exposure, this seems more like any other automation and augmentation journey as we’ve seen previously with RPA and other technologies. According to Steve Hill, managing partner at AI, cybersecurity, and management consulting firm OakTruss Group, the adoption problems aren’t technical. “RPA taught us that people matter in ways that were unforeseen,” he says. “Agentic AI is the new bandwagon, but many failures will come from lack of attention to culture, change management , workforce trust, and clarity of purpose, not the models themselves.” CIOs should therefore expect AI transformations to drag on just like digital transformations . Purchasing the tech will be the easy part, but fundamentally rethinking and redesigning all aspects of the business — including operations, processes, and products and services —will be a heavy lift requiring more critical thinking, and by more people. Manage your AI portfolio across the full innovation lifecycle With current attention squarely on AI governance, it’s easy to focus on downstream aspects such as AI risk, compliance, trust, ethics, security, sovereignty, and sustainability. Of course, all this needs to be considered and planned well in advance, and the earlier in the innovation lifecycle, the better. As attention moves to scaling, CIOs need to ensure they maintain the tools and processes to professionally manage the front-end of the innovation lifecycle as well. While excessive AI pilots and prototypes are criticized in today’s environment, the truth is they’re still essential to maintaining a healthy and continuous innovation pipeline from idea to value. CIOs should ensure they have robust means to identify and prioritize AI-related ideas, inventory AI use cases across the enterprise, and track all their associated meta-data across finance, IT and governance, and risk and compliance. To get started on identifying and prioritizing AI-related ideas, and to make it a core competency, look to techniques such as innovation workshops as well as the software-side of things. Workshops incorporate the human-side of AI transformation by way of highly-collaborative, interactive sessions bringing in a cross-functional set of subject matter experts and stakeholders that software alone can’t replicate. Assess your team’s skills as well as your AI Just as CIOs apply governance and associated guardrails around AI, they also need to examine their teams — where do you need to retrain, upskill, and hire? It often takes more skill to work with AI, and decide when to use and not use it, than perform the work in the first place. The analogy of moving from a pyramid-shaped workforce to a diamond-shaped one can be misleading. While entry-level jobs can be replaced with AI, not all entry-level jobs are created equal. Recent MBA graduates , for example, may come into entry-level roles, but they have the business acumen and critical thinking skills the organization needs more of. Just because AI can give the impression for teams to think less, they shouldn’t. In fact, it’s important to look for team members who are self-motivated to think differently and examine AI outputs more at every step. For example, AI is notoriously bad at providing strategy advice and often produces trendslop instead. So do teams have the intelligence to analyze and interpret every AI output, and determine the signal from the noise, or do they just take it on face value and act on it? Look for individuals who don’t just sit back and propagate AI slop in their workflows, decisions, and emails, but know where and when to use it and apply their own judgment. The regular assessment of your team’s skills is as essential as the regular assessment of your AI. Automate when appropriate In addition to having teams that know where and when to rely on AI, it’s important to take a similar approach for each AI use case and application across the enterprise. Determine when you need probabilistic versus deterministic code , when you need both, and when you need human-in-the-loop or not. In high-risk AI situations, you may decide to prohibit the deployment of autonomous systems in core financial or customer-facing workflows unless the underlying model and its orchestration layer have successfully passed a pilot with documented safety metrics. As reported previously in KPMG’s Q1 2026 AI Pulse Survey , these types of restrictions are well underway, with 43% of organizations identifying high-risk use cases where autonomous agent decision-making isn’t allowed. Overall, success in AI transformation is less about eliminating jobs and more about carefully rethinking and redesigning how work gets done, including where and when to use human skills and AI, and more often, where and when to carefully orchestrate both. The humans you plug into this new AI transformation need to be smarter than ever.
Score: 15🌐 MovesJun 1, 2026https://www.cio.com/article/4177613/4-recs-for-cios-to-stay-on-the-human-side-of-ai-transformation.html - The Day the Internet Is No Longer Built for Humans: When AI Tokens Become the ‘New Oil’.
On May 28, 2026, I opened my laptop and saw three headlines that rewired my brain. I don’t mean that metaphorically. I read them, closed all my tabs, stared at the wall for about five minutes, and then pulled up a blank document. Three simultaneous announcements reveal AI tokens are becoming a tradeable commodity, the internet is being rebuilt for machine-to-machine traffic, and agents are getting their own payment rails. It’s a structural shift. The first one: Reuters reported that China’s Shanghai Futures Exchange is designing a derivatives market for AI tokens [Source]. The second: Cloudflare told TechCrunch that non-human internet traffic will exceed human traffic by the first half of 2027 — bots already account for 31% of all HTTP requests [Source]. The third: Visa announced an investment in Replit, accompanied by something called the “Trusted Agent Protocol” — a system that lets AI agents verify their identity and complete payments autonomously [Source]. Individually, any one of these is a solid news day. Together, they told me something I hadn’t fully processed: the digital economy is being restructured from the ground up, and almost nobody outside of infrastructure teams is paying attention. Tokens Are Becoming a Commodity — And Wall Street Noticed Here’s a number that stopped me cold: DeepSeek V4 Pro charges $0.003625 per million tokens for cache reads. OpenAI’s GPT-5.5 and Anthropic’s Claude Sonnet? Over $0.31 per million on the same metric. That’s an 87x gap [Source]. And here’s the punchline: 80 to 90 percent of tokens consumed by real-world AI agents are cache-read tokens, according to Val Bercovici, Chief AI Officer at WEKA, a company that provides high-speed storage for exactly this kind of workload [Source]. The reason isn’t just pricing strategy. DeepSeek’s architecture is genuinely built differently. They introduced Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to slash KV-cache usage by 90 percent across their 1-million-token context window. They use Multi-head Latent Attention (MLA) to offload heavy data payloads from GPU memory into cheaper storage tiers. A 1.6-trillion-parameter model needs just 5.48 GB of high-bandwidth memory to hold a million-token context loop. A comparable Western architecture chokes at 89 GB for the same load. I started this piece thinking the story was about Chinese AI catching up. I ended up finding something stranger: DeepSeek’s architecture was shaped by U.S. export controls that cut them off from Nvidia’s best GPUs. They didn’t just find a workaround — the workaround produced a cost structure so superior that it’s now forcing the entire industry to rethink how tokens are priced, served, and traded. And “traded” is no longer a metaphor. The Shanghai Futures Exchange is designing derivatives contracts for AI tokens. CME Group — yes, the Chicago Mercantile Exchange — is working on GPU compute futures. ICE, the company that owns the New York Stock Exchange, is doing the same [Source]. Think about what that means. A token — the atomic unit of AI computation — is being transformed from a metered service into a tradeable financial asset. Businesses will be able to hedge their AI costs the way airlines hedge jet fuel. Hedge funds will be able to go long on GPT-5.5’s API price while shorting open-source alternatives. This is electrification all over again. Electricity wasn’t a commodity when Edison built Pearl Street Station. It became one once the grid standardized it and futures markets gave it a price discovery mechanism. AI tokens are on the exact same trajectory, compressed into two years instead of fifty. Fig 1: API Token Pricing — Cache Read Cost per Million Tokens (USD)$0.0036DeepSeek V4 Pro$0.87/M outputDeepSeek$30/M outputGPT-5.5$15/M outputClaude Sonnet87x gap80–90% of agent tokens are cache reads — DeepSeek’s architecture targets exactly thisCSA/HCA compression + MLA memory offloading = 87x cheaper on the metric that matters mostSources: DeepSeek V4 Technical Report, VentureBeat, OpenRouter rankings — May 20267–17x cheaper87x on cacheMIT license The Internet Is Being Rebuilt for Machines, Not Humans I remember sitting in a conference room in 2019, listening to a cloud architect explain why they needed reservation-based pricing for server instances. The reasoning was straightforward: human traffic is predictable. People sleep. People have work hours. You can forecast demand six months out. AI agents violate every assumption in that model. They don’t sleep. They don’t request one thing at a time. A single agent task can spawn dozens of sub-agents, each querying hundreds of databases, calling dozens of APIs, and reading megabytes of context — all within seconds, and then vanishing without a trace. Tia White, the general manager for Amazon OpenSearch Service, put it plainly to TechCrunch: “They spike without warning, they go idle without notice, and enterprise needs search that keeps up without paying for empty or idle compute” [Source]. AWS’s response was to decouple compute from storage in OpenSearch Serverless. The old architecture meant you always had at least one instance running — like paying for a parking spot whether or not your car is there. The new one scales compute up in seconds when an agent triggers a task and back down to zero when it’s done. You pay for exactly what you use, down to the second. Cloudflare’s numbers make this concrete: bots accounted for 31 percent of all HTTP traffic in the last six months. AI crawlers, search engines, and assistants made up roughly a quarter of all bot requests. Cloudflare senior product manager Lai Yi Ohlsen told TechCrunch point-blank: “Non-human traffic will exceed human traffic sometime in the first half of 2027” [Source]. The same pattern is playing out across the industry. Databricks and Snowflake are repositioning as AI memory and retrieval systems. Microsoft Azure is shipping updates specifically for agent traffic bursts and inter-agent memory sharing. Cloudflare launched Agent Cloud. The internet’s plumbing — originally designed for a world where one human makes one request and gets one response — is being ripped out and replaced with infrastructure that treats agents as first-class citizens. Fig 2: Non-Human Internet Traffic Trajectory20232024202520262027HumanNon-Human ↗2026: 31%2027H1: >50%Infrastructure rebuild:AWS OpenSearch Serverless (compute/storage decoupling) | Cloudflare Agent Cloud | Azure Agent-aware updates | Databricks/Snowflake repositionSources: Cloudflare Radar, TechCrunch, AWS Official Blog — May 2026 Agents Aren’t Just Using the Internet — They’re Getting Their Own Economy This is the piece I almost missed. I read the Visa-Replit announcement and initially filed it under “corporate partnership.” Then I re-read it. Visa isn’t just investing in Replit. They’re building something called the Trusted Agent Protocol — a system that lets AI agents present verifiable credentials, declare their intent, and attach relevant customer information when making transactions [Source]. That’s not a payments integration. That’s a digital identity layer for non-human economic actors. And it’s not just Visa. That same week, Robinhood launched agent-powered stock trading. Google announced an AI-powered universal shopping cart that follows your journey across the internet. Stripe deepened its partnership with OpenRouter, the token aggregator that just raised $113 million from Snowflake Ventures, Databricks Ventures, Nvidia’s NVentures, and Google’s CapitalG [Source]. I started seeing the architecture: a three-layer stack forming in real time. The bottom layer is tokens — becoming commoditized, priced, and traded. The middle layer is the network — being rebuilt to handle agent traffic patterns. The top layer is commerce — agents getting identity, payment rails, and the ability to transact. Every layer reinforces the others. Cheaper tokens mean more agents. More agents mean more pressure to rebuild infrastructure. Better infrastructure means more use cases for agent payments. More agent payments create more demand for token derivatives as hedging instruments. It’s a flywheel, and it’s already spinning. I don’t think I’m overstating this. The last time three layers of the digital economy restructured simultaneously was the early 1990s — when telecom minutes were deregulated, fiber networks were laid, and billing systems standardized. That restructuring created the substrate on which the consumer internet was built. What’s happening now with tokens, agent-native infrastructure, and agent payments is the 1990s all over again, except the primary user isn’t a person clicking a link — it’s an agent executing a multi-step workflow. What I Don’t Know I started this thinking I’d found a clean narrative about AI infrastructure investing. I ended up finding three open questions that genuinely trouble me. First: can you actually standardize a token futures contract? Oil works because WTI crude is WTI crude. But a GPT-5.5 output token and a DeepSeek V4 Pro output token represent fundamentally different units of “intelligence.” How do you write a derivatives contract when the underlying asset’s quality is non-fungible? The exchanges are clearly working on this — but I haven’t seen a clean answer yet. Second: there’s a tension between the growth of agent traffic and the collapse of unit pricing. If DeepSeek keeps driving cache-read prices toward zero, and if infrastructure companies are spending billions to handle more agent traffic — at some point the revenue model for “agent-native cloud” has to come from somewhere other than volume. Nobody has figured out where yet. Third: geopolitics. DeepSeek’s 87x cost advantage is, at its root, a product of U.S. export controls that forced their engineers to build around hardware constraints. If sanctions tighten, or if Western compliance boards rule DeepSeek’s MIT-licensed weights off-limits, the entire pricing floor shifts again. The token derivatives market would have to price in not just supply and demand, but the risk of a model being sanctioned. A token futures contract that has to account for geopolitics. I’ll be honest — I find that both fascinating and unnerving. I spent most of my career watching software eat the world. This is different. This is the world rebuilding itself so software can have its own economy, its own infrastructure, and its own financial instruments. The biggest question isn’t whether it’s happening. It’s whether anyone is building the right things for the world that’s arriving. Sources Ram Iyer — “Just like gold and oil, we’ll soon be able to trade AI token futures” — https://techcrunch.com/2026/05/28/just-like-gold-and-oil-well-soon-be-able-to-trade-ai-token-futures/ Rebecca Bellan — “The internet is being rebuilt for machines” — https://techcrunch.com/2026/05/28/the-internet-is-being-rebuilt-for-machines/ Ivan Mehta — “Visa invests in Replit to power agentic payments” — https://techcrunch.com/2026/05/28/visa-invests-in-replit-to-power-agentic-payments-for-developers/ Matt Marshall — “How DeepSeek’s radical architecture is shattering Silicon Valley’s token moat” — https://venturebeat.com/infrastructure/how-deepseeks-radical-architecture-is-shattering-silicon-valleys-token-moat/ Reuters — “China works on AI token futures market, sources say” — https://www.reuters.com/world/china/china-works-ai-token-futures-market-sources-say-race-with-us-2026-05-28/ CME Group — “CME Group and Silicon Data partner to launch first compute futures” — https://www.cmegroup.com/media-room/press-releases/2026/5/12/cme_group_and_silicondatapartnertolaunchfirstcomputefutures.html ICE — “ICE and Ornn to Launch GPU Compute Futures Contracts” — https://ir.theice.com/press/news-details/2026/ICE-and-Ornn-to-Launch-GPU-Compute-Futures-Contracts/default.aspx DeepSeek V4 Technical Report — https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro SiliconAngle — “OpenRouter raises $113M to bring order to enterprise AI inference routing” — https://siliconangle.com/2026/05/26/openrouter-raises-113m-bring-order-enterprise-ai-inference-routing/ Stripe — “Stripe partners with OpenRouter” — https://stripe.com/newsroom/news/openrouter-and-stripe AWS — “Introducing next-gen OpenSearch Serverless for agentic AI” — https://aws.amazon.com/blogs/aws/introducing-the-next-generation-of-amazon-opensearch-serverless-for-building-your-agentic-ai-applications/ The Day the Internet Is No Longer Built for Humans: When AI Tokens Become the ‘New Oil’. was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- Managing the Complexities of AI Adoption
Managing the Complexities of AI Adoption CMU Software Engineering Institute