THE TECH BRIEF, BY MICHAI MEDIA

While Chicago and Kansas City get the headlines, St. Louis has been quietly becoming one of the Midwest's more active AI hiring markets. In the last 60 days, Boeing's St. Louis defense division posted 23 AI/ML engineering roles, Centene Corporation added 14 machine learning positions in their healthcare data division, and Edward Jones launched an internal AI task force that's already absorbing talent from local universities.

More telling: three St. Louis-based regional banks, Midwest BankCentre, Heartland Bank, and Royal Banks of Missouri, all posted AI product manager roles in February. Banks don't hire AI product managers unless they're building AI products. The question is whether they're building those products to serve customers better, or to eliminate roles. Probably both.

Also worth watching: Cortex Innovation Community announced a new AI Launchpad cohort starting in May. Applications open April 7th. If you know a founder building in this space, send them the link.

The insurance industry isn't a flashy AI story. There are no viral demos, no breathless TechCrunch coverage. But it may be the most aggressively automating sector in the American economy right now, and it has direct implications for St. Louis, where insurance and financial services employ over 45,000 people.

Here's what's happening: claims processing, the act of reviewing a claim, verifying coverage, calculating a payout, and issuing a check, is one of the most document-intensive, repetitive, judgment-light jobs in the industry. AI systems can now handle the full workflow for straightforward claims (auto, property damage, medical billing) with accuracy rates that match or exceed junior adjusters. Lemonade Insurance processes over 30% of its claims with zero human involvement. Allstate, State Farm, and USAA are all quietly piloting similar systems.

What does this mean? Insurance companies will need fewer claims adjusters, fewer processors, and fewer back-office staff. But they'll need more people who can manage AI systems, audit AI decisions, and handle the complex claims that AI can't resolve. The shape of the workforce changes, it doesn't just shrink.

The most important nuance: AI claims systems make systematic errors at scale. One misconfigured rule can incorrectly deny thousands of claims simultaneously. The litigation opportunity for attorneys who understand AI decision-making is significant and largely untapped.

Most people use AI writing tools like a fancier Google search, type in a request, get back something generic, spend 45 minutes editing it into shape. Here's a prompt structure that eliminates most of that editing time by giving the AI everything it needs to actually match your voice and purpose.

It's called the Before / After / Bridge framework. Here's the structure:

We've tested this across client email campaigns, landing page copy, and internal SOPs. Compared to a generic "write me an email about X" prompt, the Before/After/Bridge structure reduces editing time by roughly 60% because the AI understands what the content is supposed to accomplish, not just what it's about.

This week's action: Pick one email you have to write this week. Use the BAB structure. Time yourself editing the output versus your usual process.

It was another week of major model releases, OpenAI dropped updates to their reasoning models, Google pushed Gemini Ultra into more enterprise tools, and Anthropic quietly shipped improved Claude capabilities for document analysis. The tech press treated each as a seismic event. The reality for St. Louis business owners is more nuanced.

Here's the actual signal underneath the noise: the cost of AI inference is dropping at a pace that's faster than most people realize. Running AI on a task that cost $0.30 per execution 18 months ago now costs closer to $0.02. That's not a marketing claim, it's the result of model efficiency improvements and infrastructure competition. The practical consequence is that automation projects that weren't economically viable two years ago are now clear no-brainers.

Specifically: if you looked at AI-powered customer support, document processing, or voice automation in 2024 and decided the ROI wasn't there, you should revisit that math. The cost side of the equation has moved significantly in your favor.

Marcus Webb is a 22-year-old computer science senior at Washington University in St. Louis. Last spring, he took a part-time job at a small physical therapy clinic in Clayton, mostly to cover rent. Within two weeks, he noticed that the front desk staff spent about four hours every day on the phone managing appointment scheduling, confirmations, and reschedules. Manual, repetitive, and entirely predictable.

He spent six weeks building a voice AI scheduling system using open-source components, the clinic's existing scheduling software API, and a custom voice model. He deployed it with the clinic owner's permission in August. By October, the clinic had reduced front desk scheduling calls by 68%, and the owner, impressed enough to pay Marcus a real consulting fee, introduced him to five colleagues.

Today, Marcus's system runs in 12 clinics across St. Louis, St. Charles, and Jefferson County. He's doing this between classes. He's not looking for VC funding. He's not trying to be the next startup unicorn. He's solving a specific problem for a specific type of business in his city, and it works.

If you want to connect with Marcus or know a student doing something similar worth spotlighting, reply to this email. We read every response.