2 stories · 1 sources

Artificial Intelligence

MIT Technology Review

MIT Technology Review: Rethinking organizational design in the age of agentic AI

MIT Technology Review: The Download: puncturing the AI jobs panic

MIT Technology Review highlights a widening gap between the appetite for AI agents and the ability to deploy them cleanly. In one cited survey, 85% of organizations said they want to be agentic within three years, while 76% said their current operations and infrastructure can’t support that shift. The core problem is not just tooling; it is that many firms are layering agents onto human-centered workflows instead of rewiring the work itself.

That matters because agentic AI is being framed as something that can execute entire workflows with limited human input, not just draft text or summarize data. The article argues that early gains may be real in customer service, HR, and sales, but the bigger gains will only show up if businesses redesign processes, roles, and controls around the technology rather than treating it like a software patch.

Takeaway:

• AI agents are becoming an operating-model issue, not just a product rollout.
• Firms that keep adding agents to broken workflows may cap the upside and invite disappointment.

Source: https://www.technologyreview.com/2026/05/26/1137584/rethinking-organizational-design-in-the-age-of-agentic-ai/

MIT Technology Review also pushes back on the idea that AI is already wiping out white-collar jobs at scale. The piece says analysis of US labor data shows unemployment is actually lower in occupations most exposed to AI than in less-exposed ones, and it finds no sign of a mass shift from AI-exposed work into manual labor. The headline story is not mass unemployment; it is that the data do not yet support the loudest versions of the panic.

But the more useful warning is narrower: entry-level work may be getting squeezed first. A Stanford study cited in the source found that workers aged 22 to 25 in the most AI-exposed occupations saw a 16% relative decline in employment after generative AI spread, while more experienced workers in those same jobs did not see the same drop. For operators, that suggests the near-term risk is not a generic “AI takes all jobs” scenario, but a thinning of junior hiring pipelines in specific functions.

Takeaway:

• The broad labor-market story is still mixed, not catastrophic.
• The early warning signal is concentrated in junior roles where AI can absorb routine starter tasks.

Source: https://www.technologyreview.com/2026/05/26/1137865/its-time-to-address-the-looming-crisis-in-entry-level-work/

2 stories · 4 sources

Cryptocurrency

CoinDesk · The Block · Decrypt · Unchained

CoinDesk: Crypto PACs spend $9 million in Texas and score wins in both parties

The Block: China’s top court to study judicial rules for crypto amid rise in related cases

CoinDesk reports that crypto-focused PACs spent more than $9 million in Texas this cycle and backed winning candidates in both parties. The spending helped deliver primary victories for industry-aligned candidates, including Democrats and Republicans, and the piece frames this as evidence that digital assets have become a cross-party electoral force rather than a one-party hobby.

For crypto operators, the important detail is not horse-race politics; it is the infrastructure of influence. The article suggests industry groups are now organized enough to target races, support preferred candidates, and shape the policy environment ahead of the 2026 midterms. That raises the stakes for compliance, lobbying strategy, and public-policy planning around custody, payments, and regulation.

Takeaway:

• Crypto’s political footprint is no longer confined to one side of the aisle.
• The Texas spending shows the industry is building a repeatable policy influence model, not just making symbolic donations.

Source: https://www.coindesk.com/policy/2026/05/27/crypto-pacs-spend-usd9-million-in-texas-and-score-wins-in-both-parties

Unchained reports a supply-chain malware campaign called TrapDoor that targeted crypto developer environments tied to Aptos, Sui, and Solana. Socket Security identified more than 34 malicious packages across npm, PyPI, and Crates.io, with install-time triggers that could fire automatically during normal development workflows. The exfiltration targets included SSH keys, wallet keystores, AWS credentials, GitHub tokens, browser data, environment variables, API keys, and local config files.

This is a security story, but it is also a product-story for teams building crypto infrastructure. If malicious packages can blend into ordinary installs and builds, then dependency hygiene, registry monitoring, credential scoping, and developer-environment isolation become core operational controls rather than optional hardening steps.

Takeaway:

• The attack surface is the software supply chain itself, not just the wallet or the chain.
• Crypto dev teams should treat package trust, secrets management, and build isolation as frontline security issues.

Source: https://unchainedcrypto.com/trapdoor-malware-campaign-targets-crypto-developer-environments-with-34-malicious-packages/

2 stories · 4 sources

Data Science

NVIDIA Technical Blog · Towards Data Science · KDnuggets · Databricks Blog

NVIDIA Technical Blog: Extract More Kernel Performance with NVIDIA CompileIQ Auto-Tuning

Towards Data Science: What Is a Data Agent?

NVIDIA’s Technical Blog says CUDA 13.3 includes CompileIQ, an AI-powered compiler auto-tuning framework that uses evolutionary and genetic algorithms to optimize compiler parameters for specific GPU workloads. The pitch is straightforward: after teams have already tuned batch sizes, quantization, flash attention, and kernel fusion, they can still search for gains by making the compiler itself tunable.

For data science and ML infrastructure teams, the significance is less about novelty and more about operational leverage. The tool is aimed at workloads where small kernel hotspots dominate total compute, such as LLM inference, and it is designed to balance runtime, compile time, and power consumption while producing reproducible configurations. That makes it relevant to anyone chasing throughput without blowing up deployment consistency.

Takeaway:

• Compiler behavior is becoming another tuning surface in GPU-based AI systems.
• The practical value is in squeezing incremental performance from hotspots that dominate end-to-end throughput.

Source: https://developer.nvidia.com/blog/extract-more-kernel-performance-with-nvidia-compileiq-auto-tuning/

A Towards Data Science post describes a data agent as “a report you can talk to,” with Microsoft Fabric’s data agent as the example. The article argues that analysts spend too much time building visualizations and answering ad hoc KPI questions, and that a governed data agent could let business users ask for insights directly instead of waiting for a new dashboard or spreadsheet.

The useful takeaway is not that dashboards are dead—they are not—but that the interaction model is shifting. If the reporting layer can be wrapped in prompts against governed data estates, then analytics teams may spend less time formatting charts and more time managing semantics, trust, and access to the underlying data model.

Takeaway:

• Data agents are being framed as an interface shift for analytics, not a replacement for governed data.
• The operational challenge moves from chart-building toward semantic consistency and data access design.

Source: https://towardsdatascience.com/what-is-a-data-agent/