profile

mAIn Street

mAIn Street #307: Clinicians are warming up to AI for explaining things to their patients; Microsoft preps huge security update to Windows after AI callouts; A tool for scheduled analyses

Published about 17 hours ago • 8 min read
AI explains test results, changes daily work and brings media generation into the editing timeline.
mAIn Street Daily Newsletter. AI news for people who actually have jobs to do.
Tuesday, September 8, 2026 · Edition 307
Today’s throughline
AI is drafting patient notes, preparing claims and handling more of the routine work that slows teams down.
Today’s edition follows hospital results, workplace habits, software fixes and tools moving into daily operations.
Top 5
Five stories to start your day.
01
Epic’s Art drafts patient-friendly explanations of lab and imaging results for clinicians to review and edit. Memorial Hermann reported 32% fewer follow-up messages when clinicians started with those drafts, offering a concrete example of AI improving routine patient communication.
02
The Verge reported that more than 650 Windows security fixes were expected in Tuesday’s release. That was a preview of the update, with AI-assisted vulnerability discovery contributing to the growing patch workload.
Source: The Verge
03
Among 306 senior leaders at large U.S. companies already using AI, 66% reported productivity improvements. Only 45% described an established practice for continually reviewing how work should change, leaving managers with more to do after buying the tools.
04
Editors can select a timeline gap and generate video, music or sound effects there. The update reduces the need to move between an editing project and separate generation tools.
Source: The Verge
05
More than 400 subscriptions will support interested graduate students and faculty at CUNY and Northwestern. The program gives future journalists access to practice with AI as news organizations work through questions about reporting, verification and editorial judgment.
Source: OpenAI
Adobe Premiere generating a clock-ticking sound effect in the editing timeline.
Premiere’s new media-generation controls. Image: Adobe, via The Verge.

AI Workflows | Tuesday: Context and harness management

Working backward helps you check whether AI understood the assignment.

Google's general counsel emphasized human judgment in today's Axios interview.

I have a practical way to bring my own judgment back into an AI project: start at the end.

For something I'm writing, I'll read the last sentence first, then the sentence before it. Then I'll read the final paragraph, followed by the next-to-last paragraph, and keep working backward.

Start at the end.

1. Last sentence
2. Previous sentence
3. Final paragraph
4. Previous paragraph

I need that time outside AI to get my own thinking going again. This helps me see whether AI understood the mission. It also helps me reinsert critical thinking into my handling of the finished work.

Try it with a customer proposal. Suppose your assignment was to recommend a small pilot. The closing paragraph asks the customer to commit to a companywide rollout.

Start there. Does that request match the decision you wanted the customer to make? Work backward through the proposal and examine what supports it. You might discover that several sections assume a commitment the customer hasn't made.

Now you have a specific correction: bring the proposal back to the pilot. You can also identify material worth preserving, such as the approved pricing and the customer's stated problem.

That review gives you something useful to carry into Tuesday's skill: context and harness management.

Context is the information AI needs to do the job, including the goal, source material, and decisions you've already made. The harness is the working setup: standing instructions, connected sources, available tools, and the points where you review the work.

Make your review part of that setup.

When you return to AI, have it turn your findings into a short, updated project brief. It should state the intended outcome, what remains approved, and what needs to change. Read that brief yourself before authorizing another revision.

For example, ChatGPT Projects can keep related chats, files, instructions, and connected sources together. Have AI use the material already available in your project to prepare the brief. Save the approved response through its message menu using Save to project or Add to project sources.

Here's an assignment you can adapt after your independent review:

Use the current proposal, approved source material, and my review notes to prepare an updated project brief. The goal is a small pilot. Preserve the approved pricing and the customer's stated problem. Identify every passage that assumes a companywide rollout. Show me any conflicts or missing information before revising. After I approve the brief, revise the proposal and provide a short change log. Keep the previous version available. The revised proposal is for my review before it goes to the customer.

On the next revision, ask AI to consult that approved brief and flag conflicting instructions. Afterward, read the document forward to check the full argument. Verify factual claims against their sources.

For a manager or business owner, make the review concrete. Ask the person responsible for the work to explain what they changed and why. Keep the approved decisions available with the project so the next person can understand them, too.

Take one AI-assisted document today and read its final sentence first. Before returning to the chat, write down what that sentence needs to accomplish.

AI Tool Spotlight
Put a recurring report on a schedule.
Databox Routines runs a saved analysis on a daily, weekly, monthly or custom schedule, then delivers the result through email, Slack or the app. Previous runs stay available, and you can ask follow-up questions about a report.
A useful first job: review last week’s website traffic and lead conversions before the team meeting. Connect the relevant data, define what counts as a lead and save the analysis instructions. Check the first report against your source systems before scheduling it.
Best fit: managers, agencies and business owners who rebuild the same performance report every week.
Headlines
And that was only the Top 5
Twenty-five stories shaping work, business and daily life.
Work, skills & customer service
Only 31% of surveyed leaders said AI development for employees was part of broader workforce planning. Ipsos interviewed 306 directors and executives at U.S. companies with established AI adoption and at least $100 million in annual revenue.
The 2026–27 support involves CUNY’s Newmark journalism school and Northwestern’s Medill School. OpenAI also renewed its work with the American Journalism Project to support local newsrooms.
The collaboration includes assistants for payroll anomalies, reports and employee questions, plus agents designed to act under human oversight. ADP is also using AWS tools to modernize its tax and payroll applications.
Its Assist API supplies knowledge gathered by an agent that works through a product’s actual screens and workflows. Software companies can use it to answer questions or guide customers through setup, with support teams able to review and correct responses.
Ana develops a negotiation strategy, drafts vendor exchanges and tracks offers using Vertice’s purchasing data. Teams can choose autonomous negotiation or review each round themselves, and can adjust price, payment and contract priorities along the way.
Health & science
Memorial Hermann reported 32% fewer follow-up questions when clinicians used Epic’s draft explanations, which they review and personalize before sending. The health system’s results reflect its own rollout.
The VELMENI integration lets Isaac PracticeOS assemble supporting images and documentation after a dentist confirms a finding. The company requires review by a dentist and billing specialist before a claim is submitted.
The update gathers chart findings, echocardiograms and ECG information on one patient screen and adds lists organized by disease severity. It aims to make relevant information easier to find across the more than 80 U.S. cardiovascular programs using the platform.
AlphaGenome Atlas predicts the molecular effects of possible single-letter DNA changes and is free for academic research. Scientists can use it to prioritize variants for further investigation; its predictions still need biological validation.
Security, trust & public policy
The anticipated volume adds to the workload for IT teams reviewing, testing and deploying security updates.
The Tech Transparency Project identified more than 300 ads containing child sexual abuse material, often created with AI. It reported that Meta removed about 150 still-visible ads after receiving its findings, then continued to carry additional abusive ads.
The system combines existing security findings with financial context to estimate possible losses and the value of remediation. Proposed actions require human approval, giving security and finance teams a common basis for deciding what to fix first.
The available update explains vulnerability workarounds and helps turn device commands into reusable automations. Network teams can inspect and adjust the proposed changes before running them.
The new Caspian integration ranks weaknesses using the operator’s equipment, threats and risks, then prepares remediation plans. It is available for teams protecting operational technology, where taking equipment offline can carry serious consequences.
Draft rules would also extend child-safety duties to AI chatbots.
Tools & daily operations
New audio features and an After Effects assistant are also entering beta, expanding the tasks editors can handle within their existing projects.
Its Listing Optimizer scores listings and recommends changes to photos, pricing visibility and missing details. New customer examples point to better engagement, though the company’s before-and-after results do not establish how much of the improvement came from the tool.
Brokers and insurers can submit application documents for a summary of operations, exposures and loss history. The service offers a quicker starting point for reviewing a submission’s risks and supporting information.
The planned integration will carry loan, budget and property information into Built’s platform for construction administration and portfolio management. Banks and credit unions could spend less time reconciling spreadsheets and separate systems after a loan closes.
Ascent 3.0 includes the Venn assistant and a connection for customers’ own AI tools. Changes that could delete or materially alter configurations require administrator approval; the release is available within existing subscription tiers.
Expected in the fourth quarter, the toolkit will let selected modules analyze sensor data without an extra AI processor. Potential applications include spotting machinery problems and reading existing analog meters with connected cameras.
In its Braid demonstration, a model on another computer incorporated a shared spending limit into its answer. New topic-based Channels organize these exchanges for people experimenting with multiple models on their own machines.
AI research & infrastructure
The collaboration covers customized chips for running AI and faster optical connections inside data centers. Qualcomm also plans to expand its use of AWS AI infrastructure to shorten chip-design work.
Its September 6 report counted 3.1 agent-workdays of runtime per human workday by mid-August. That measures machine activity, not an equivalent productivity gain, and the company reported substantial human intervention on longer tasks.
In a September 6 essay, Jakub Pachocki argued for enforceable safety thresholds and voluntary slowdowns as models become harder to monitor. His proposal would give outside auditors or public authorities a larger role in overseeing continued development.
mAIn Street gives nontechnical readers useful AI news they can put to work.
mAIn Street #307 · Tuesday, September 8, 2026

​Read All mAIn Street Back Issues Here​

mAIn Street

Aric Mitchell

Check out the resources I offer below and sign up for my new newsletter!

Read more from mAIn Street
A house visualization built with Astra and Blender, showing an open living area beside a garden and pool.

Minnesota’s AI-image law stays in force, lawmakers propose school AI labs, and Docusign plans a September 30 connection to major assistants. Monday, September 7, 2026 · Edition 306 Today’s throughline AI is bringing new decisions to classrooms, courtrooms and everyday business. This Labor Day edition follows school funding proposals, image protections, contract tools and support for the teams defending essential services. Top 5 Five stories to start your week. 01 A judge left Minnesota’s ban...

2 days ago • 9 min read
A professional reviewing work on a laptop in an office

OpenAI began Astra’s staged rollout, Sanders and Casar announced a bill to ban artificial superintelligence, and three major AI services went down together. Friday, September 4, 2026 · Edition 305 Today’s throughline AI’s next capability leap arrived alongside a congressional demand to stop it. Astra’s staged rollout, a proposed superintelligence ban and Thursday’s outages put capability, control and continuity into the same morning briefing. Top 5 What matters today 01 OpenAI began Astra’s...

5 days ago • 9 min read
Path to Astra: Frontier capabilities and safeguards.

OpenAI says Astra reached a Critical cyber threshold, Google released Gemini 3.8 Flash, and New York City paused classroom AI through eighth grade. Thursday, September 3, 2026 · Edition 304 Today’s throughline More capable models are forcing institutions to define access, verification and accountability. Today’s stories move from frontier cyber tools to classrooms, courts and everyday work, with control and verification at the center. Top 5 What matters today 01 OpenAI said Astra is its first...

6 days ago • 9 min read