More Human AI · Customer Center of Excellence · by Strategy For AIPublic reading room
More Human AI
What is this?

A center of excellence for AI in mid-market firms.

Every 48 hours, Brad Bush takes what crossed his desk, runs it through Strategy For AI's frameworks, and publishes one brief. The latest is below — free. Subscribers get the library, search, and Brad's commentary on every source.

Brief №0 · Sep 14

Your Associates Are Losing Twenty Days a Year Babysitting AI Drafts

Mid-sized professional-services leaders risk eroding their junior talent pipeline as associates quietly absorb weeks of prompt troubleshooting and unchecked error review.

Executive briefing · 30-second readTHE ARGUMENT
01Enterprise technology leaders report sweeping productivity gains while desk workers lose four full workweeks each year troubleshooting algorithmic drafts.
02Automating routine entry tasks destroys junior skill formation and traps future firm partners in the Autopilot Paradox.
03Replace broad license rollouts with targeted AI Impact Projects that enforce deliberate partner verification gates on complex client deliverables.

Late Thursday afternoon, walk down the hall of your 150-person firm and look into a second-year associate's office. You will see an ambitious professional staring at two monitors with tired eyes. Complex statutory tax regulations and client advisory memos sit untouched on the desk. Instead, they are rewriting a prompt for the fifth time, trying to coax a language model into extracting cap table figures without inventing share classes. The tension in their shoulders is unmistakable. Their billable target never moved, yet their entire afternoon vanished into troubleshooting software.

Executive briefings landing on your desk paint a picture of effortless speed. Azeem Azhar surveyed 250 enterprise technology leaders and found that 95% now report meaningful results from artificial intelligence deployments. Software vendors are posting staggering growth. Legora raised capital after three founders with zero legal training built a platform that captured over 3% of the world's practicing lawyers, hitting $100 million in annual recurring revenue in just 18 months.

SRC [5]Legora's 6-month sales freeze after a $35M raiseThe AI Corner
Legora launched with 3 founders and 0 lawyers. 3 years later, over 3% of the world’s lawyers use the product they built anyway.

Vendor sales decks promise immediate efficiency, while the floor reality tells a different story. Marcus Sawyerr examined new data from BambooHR surveying 1,608 salaried desk workers. Employees now spend roughly 20 eight-hour workdays every single year troubleshooting machine outputs and iterating on prompts. That is four full business weeks burned fixing formatting errors, validating hallucinated case numbers, and coaxing models into usable shape. The administrative burden did not disappear. It migrated down to your newest hires.

SRC [7]AI workers should remove admin, not move itMarcus Sawyerr (Substack)
BambooHR’s new research estimates that workers spend roughly 20 eight-hour workdays a year troubleshooting AI and iterating on prompts. A recruiter spending less time in the ATS but more time correcting an assistant has not necessarily got their day back.

Picture an apprentice's workbench. For a hundred years in legal, accounting, and advisory firms, senior partners learned their craft at that wooden table. You shaved rough lumber with a hand plane. You ran manual footings on balance sheets. You read hundreds of messy lease agreements to draft your first standard indemnification clause. That grind was slow. Yet those repetitions taught your hands the grain of the wood. You learned where deal terms break, where clients bury liability, and where financial statements disguise trouble.

Flooding that workbench with pre-cut digital drafts breaks the transmission of judgment. Andrew Ng, who has taught machine learning to 8 million students, issued a stark warning: current AI usage patterns actively disrupt how professionals learn. When juniors rely on algorithms for initial analysis, they skip the cognitive struggle where professional intuition takes root.

SRC [10]AI Does 30 to 40% of JobsThe AI Corner
Andrew Ng has taught 8 million people how machine learning works. He just told them AI is bad for learning.

This breakdown triggers what I call the Autopilot Paradox. When an automated tool performs with 90% accuracy, human vigilance plummets. Associates stop evaluating correctness and begin assuming correctness. They skim a fifty-page disclosure schedule instead of interrogating the representations. Over time, critical faculties decay. When a bizarre regulatory edge case eventually strikes a client engagement, the algorithm fails, and the firm demands senior judgment from an associate who spent their formative years acting as a software babysitter. Elite professional services cannot survive that void.

Retreating to paper files will fail, and letting software run unchecked will ruin your practice. Mid-market leaders must actively manage how technical leverage intersects with career development.

This quarter, select a single practice group and launch one focused AI Impact Project rather than scattering licenses across the firm. Establish an explicit human verification checkpoint for every automated task. Enforce a deliberate friction policy: require junior staff to write their own outline and identify three core transaction risks before consulting an AI tool. Require partners to review those initial human notes alongside the machine draft. Make the learning visible.

The wooden workbench remains the foundation of your firm. Protect the manual repetitions that build future equity partners. Strip away the wasted hours of prompt troubleshooting and rebuild the habit of disciplined judgment. The software will keep accelerating. Whether you build true craftsmen or exhausted prompt editors is up to you.

The rest of this brief examines how the conversation should be opened — what specifically to say to clients who are silent, how to structure the disclosure so it lands as competence rather than panic, and three ways the Friction Audit reveals exactly which client conversations to have first…

Continue reading · free
The rest of this brief, the library, and Brad's commentary on every source.
Join free →Log in
What we stand for

Five tenets.

Non-negotiable beliefs underneath every brief, every answer, and every recommendation. They change on the scale of decades, not news cycles.

  1. Human value comes before AI value.
  2. Humans and AI together beat either alone.
  3. Adoption is a people problem, not an engineering one.
  4. AI informs judgment; it never replaces it.
  5. Measure the whole return (to your people, your clients, and society) not just the balance sheet.

The archive · 13 more briefs

Subscriber access
trend-governance-andSEP 8Leaders must govern the whole AI stackTrend briefreflect-draft-augAUG 15You Firm Should Have a Learning Loopreflect-draft-augAUG 1Will you meter AI like a utility or buy it like software??reflect-draft-julJUL 27Intelligence Is Getting Cheap, Judgment Is Getting Expensivereflect-draft-julJUL 21Your firm needs a trust architecture before it needs another modelreflect-draft-julJUL 19The firms that win with AI will own the learning loopreflect-draft-julJUL 17Your AI firm will be judged by its brakesreflect-draft-julJUL 11Your AI advantage is getting cheaper every weekreflect-draft-julJUL 9Is your AI rollout issue a management problem?reflect-draft-julJUL 3The firms that win AI will own the control layerreflect-draft-junJUN 21The firms that win AI will redesign work before they buy autonomytrend-governance-andJUN 11Governance is becoming the new operating system for AI-enabled workTrend brieftrend-adoption-roadmapsJUN 11AI adoption is no longer a tool-choice problem; it is an operating-model redesign problemTrend brief
Subscribe · free
The full archive, the library, and Brad's commentary on every source.