AI spending lifts U.S. manufacturing, employers resume selective hiring, and states set firmer rules for healthcare AI.
|
|
Tuesday, July 28, 2026
|
|
AI spending is showing up in U.S. factory orders, employers are selectively hiring again, and states are setting firmer rules for AI in healthcare. A new security alliance and a federal utility-cost pledge show how quickly adoption is becoming a question of jobs, patient rights, network safety, and household bills.
|
A voluntary pledge aims to keep data-center expansion from raising household utility bills. Image: AP Photo/Justin Rex.
|
|
Top 5
What matters today
|
| 01 |
Core capital-goods orders rose 0.9%, shipments jumped 1.9%, and computer and electronics orders climbed 3.1%. The figures show that corporate AI spending is now large enough to support manufacturing and broader economic growth.
|
|
| 02 |
The new measures cover insurance denials, clinical oversight, therapy chatbots and disclosure. The common thread is simple: consequential health decisions still require accountable human judgment.
|
|
| 03 |
Alphabet, Booz Allen Hamilton, ServiceNow, CSX and other companies are adding staff in selected areas. The rebound is uneven, but it undercuts the idea that AI agents can remove junior and operational roles without creating new gaps.
|
|
| 04 |
Adobe, CrowdStrike, Dell, Hugging Face and other members will share tools and practices for securing models and agents. The coalition is an early attempt to turn last week’s breach into industry-wide defenses.
|
|
| 05 |
The pledge now includes 23 governors and at least 187 companies, but it is not binding. With some forecasts warning of 15% to 40% increases in monthly utility bills by 2030, the public will judge the agreement by actual rates.
|
|
|
|
|
AI Workflows
When the model can see the problem, debugging gets easier
Screenshots and screen recordings can carry details you do not yet know how to describe.
|
BKFA in action. The game became the real-world example for today's AI Workflow on multimodal debugging.
I’ve spent much of the past week building video games. That sentence would have sounded ridiculous to me a month ago. I used to be a die-hard gamer, but life has crowded out the eight-to-10-hour commitment many modern games demand. What I still miss are the tighter Super Nintendo and Sega Genesis experiences, the ones you could finish in two to four hours and then move on with your life.
That nostalgia led me to Trick-or-Treat!, a platformer that plays great on desktop and is close to mobile-ready, save for some stubborn choppiness. It also led me to BKFA, my completely made-up Bareknuckle Fighting Association game. BKFA is much closer to what I wanted from the start. A week ago, I knew el zilcho about building video games. With GPT-5.6 Sol handling the higher-level work, I have already reached the point where I enjoy playing what I built. The useful lesson reaches far beyond games. I do not know all the vocabulary a traditional developer would use to identify a rendering hitch, physics bug, timing issue, or mobile performance problem. I can still recognize the problem when I see it. So I record it.
When a game stutters on my phone, I capture the screen. When a glove detaches from a fighter’s body, I take a screenshot. When controls work on desktop and fail on mobile, I record the whole sequence, including what I tapped and what happened next. The model receives the evidence: timing, sequence, context, and the visible symptom.
That evidence changes the quality of the diagnosis. A screen recording can reveal when the slowdown begins, what action triggers it, and whether the problem gets worse over time. A screenshot preserves layout, spacing, position, and proportion. The model can compare expected behavior with actual behavior, narrow its search inside the project, and trace the likely cause through the code.
Once the model has a grip on the issue, I ask it to sweep the entire game for the same pattern. A detached glove may expose a shared animation problem. One broken mobile control may reveal an input-handling flaw across every level. A single performance hitch may point to an effect, sound routine, or update loop that keeps repeating. Fixing the visible symptom helps. Finding every place that can recreate it is where the real value shows up.
Show, don’t tell applies to AI collaboration, too. We usually treat that phrase as writing advice. Multimodal models make it useful for broken websites, confusing forms, spreadsheet errors, slide layouts, app behavior, and any other project where the problem is easier to see than explain. Ya don’t know what ya don’t know sometimes. A good recording gives the model something concrete to work with while you are still learning the language of the task.
|
Use this with a screenshot or screen recording
Review the attached material and diagnose the problem. First, describe only what is visibly happening. Then identify the most likely causes inside the project. Fix the root cause while preserving unrelated behavior. Search the entire project for other places where the same pattern could appear, correct those instances, and explain what changed and what I should test afterward.
|
The next time you cannot find the right words, give the model the right evidence.
|
|
|
AI Tool Spotlight
Databox Genie Artifacts
Turn live business data into a shareable report by asking a plain-language question.
|
|
Databox Genie connects to live business data from more than 130 integrations and answers questions without requiring SQL. Its Artifacts feature packages the result into a shareable document with charts, analysis, and context.
A useful first test: Connect one trusted data source and request a weekly performance review that identifies the largest change, explains likely drivers, and lists the three questions management should ask next. Check metric definitions before sharing the report.
Explore Databox Genie
|
|
|
Headlines
More developments from work, public life, security, business, and the global AI race.
|
Work, jobs, and business deployment |
Reuters
Core capital-goods orders rose 0.9%, shipments jumped 1.9%, and computer and electronics orders climbed 3.1%. The figures show that corporate AI spending is now large enough to support manufacturing and broader economic growth.
WhatJobs
Alphabet, Booz Allen Hamilton, ServiceNow, CSX and other companies are adding staff in selected areas. The rebound is uneven, but it undercuts the idea that AI agents can remove junior and operational roles without creating new gaps.
Axios
The NewsGuild now has roughly 85 to 90 contracts with AI provisions, while entertainment and gaming workers have won consent and notice rules. Most U.S. workers lack the bargaining leverage behind those protections.
Axios
The platform says its agents handle more than 90% of routine customer-load touches while people step in for driver disputes and data problems. It is a clear example of AI changing back-office roles without removing every human checkpoint.
Business Wire
A Zywave survey of more than 1,400 employers found that AI adoption now affects broker selection and client retention. Human strategic guidance remains valuable, which gives brokers a practical target: faster analysis with clearer advice.
Business Wire
Ignite is designed to spot revenue opportunities across the systems auto dealers already use. The useful part is the shared context, which can keep sales outreach from ignoring service history or prior customer conversations.
PR Newswire
Most workplace adoption still sits at the task-assistance stage. The finding explains why companies can spend heavily on licenses and still struggle to show broad productivity gains.
|
|
Health, schools, and public life |
Transparency Coalition
The new measures cover insurance denials, clinical oversight, therapy chatbots and disclosure. The common thread is simple: consequential health decisions still require accountable human judgment.
AP News
The pledge now includes 23 governors and at least 187 companies, but it is not binding. With some forecasts warning of 15% to 40% increases in monthly utility bills by 2030, the public will judge the agreement by actual rates.
The Guardian
Educators raised privacy, security and human-connection concerns, while Seneca Nation residents questioned the cultural judgment behind the project. The dispute shows how quickly a technology pilot can become a trust problem.
Capitol News Illinois
The state disclosed where AI assisted the 409-page document and said people reviewed the resulting language, links and graphics. That transparency gives schools a concrete example of documenting AI use with a clear record of where people reviewed the work.
University of Phoenix
The Harris Poll survey also found that 45% of respondents could not identify clear accessibility guidance in their workplace AI policies. Tools are moving faster than the rules that tell employees how to use them inclusively.
AfroTech
The effort targets professionals, entrepreneurs, creators and civic leaders. Large-scale training promises matter most when they lead to repeatable workplace skills, not one-time exposure.
Business Wire
Bunsen can interpret research goals, plan computational work, run molecular-discovery workflows and explain results. The early-access release moves scientific agents closer to handling connected research processes that span planning, execution, and interpretation.
|
|
Security, law, and markets |
Reuters
Adobe, CrowdStrike, Dell, Hugging Face and other members will share tools and practices for securing models and agents. The coalition is an early attempt to turn last week’s breach into industry-wide defenses.
Business Wire
The system focuses on sequences of individually normal actions that become risky when an agent has broad access over time. That is the control problem companies face as agents begin changing code, using tools and handling sensitive data.
Reuters
Companies do not need to wait for a new AI statute to face scrutiny over deception, discrimination or unfair practices. Existing state law already gives regulators several ways to act.
Reuters
Policy decisions now influence data centers, chips, energy, trade and model access. Investors are pricing political risk alongside technical progress and business demand.
Reuters
Investors increased exposure to life-sciences tools, pharmaceuticals and healthcare equipment. The trade assumes that faster research and strong deal activity will produce returns, an expectation the science still has to earn.
Business Wire
The system analyzes medical notes, bills and legal records, then helps carriers prioritize casualty claims. Insurance teams will still need clear appeal paths when automated risk scores influence attention and settlement strategy.
|
|
Global headlines |
Reuters
The dispute could widen from chips and export controls into investigations of how models were trained and whether U.S. technology was taken.
Reuters
The bank is pairing technical hiring with another 100 wealth-management roles. That mix shows AI investment expanding alongside human advisory services.
Reuters
The company plans to add 250 engineering and support jobs over two years as it expands its European headquarters.
Reuters
The court fight could test police facial-recognition powers in a country with no specific law governing the technology.
|
|
mAIn Street is built for nontechnical readers who want the signal, not the sludge. |
|
|
|