Tim Rayburn Photobooth for social media The monster under your be Part 1: That Tuesday “Insert Idea, Get Software” The vague menace of the most effective kind. You can’t argue with it. You cannot prepare. We are going to flip on the lights and show you the face of the monster The fear is rational as a category Your job will change. That Tuesday does not look like what you imagine. Light 1 The Hidden Dependency Confirm the strength of- don’t minimize it Idea ——> software What is in the arrow matters a great deal! Software devs: the more you understand the business you are writing for, the stronger you will be. IDEA > domain translation > precise specification > code generation (AI strong) > domain validation > software that works (Currently, only Gemini can solve this) Give AI a ping-pong table without a net, a 1-foot border, tell it to place a maximum size TV on the table. This is a spatial understanding process. Business people do not give you a perfect specification list on the first go. AI generates HIPPA-shaped code. It cannot generate code that survives clinical reality. Algorithmically correct code can be catastrophically wrong. These need to be agent accessible: Github, azure devops, jira Light 2 Compression, Not Extinction “Will AI take my job?” Is a binary question that produces binary responses AI will compress your work into high-value time What will you do with the remaining time?  E.g.: ATM and bank tellers > hire more bank tellers! VisiCalc: more accountants than ever! “The historic record of automation is, overwhelmingly , a record of compressions rather than replacement.” Light 3 The Broken Ladder New workers are the ones that suffer AI is doing adequate work We don’t know how to train juniors anymore Rote work is meant to help juniors build judgment Seniors must train juniors Currently, the training is very much different than how the seniors learned… The ratios are going to change. Old 10:1 Early career? The path will be different. We have 2 years of not hiring juniors If we do 3 more years of this, there will be a point when you want to retire, and your replacement does not exist! Part 3: the Race Car The monster is actually a race car! We have a history of faster and faster models. We are already writing C (which compiles to Assembly) We have 3rd gen langueages that compile to byte-code, and then we had larger problems to solve The question now: what problems will we be solving in 5 years that we cannot solve today. There will be a massive pick up in productivity, if you know how to drive the car! Get in and Drive! Driver: Strategy, judgment, know the track, know when to brake Mechanic: Understanding how it works, tuning for conditions when we started to electrify the world, there was no stopping that ball from rolling! Ai is rolling out. It is going to be everywhere. Infrastructure Harness - scaffolding that wraps the language model Evaluation - the test suite for AI behavior. Pass/fail signals. Prompts - which are specifications. Developers write sepcifications Spec as Strategy Well-written: Speed Amplifies Stake Part 4: The 201 Skills AI is not a tool. It is a management skill. The Six 201-Level Skills Context assembly Quality judgment - verify AI output Task decomposition - defining task composition Iterative refinement - iterate a better prompt, not just the output Workflow integration - this is how work actually gets done, not a side activity Frontier recognition - knowing when you are operating outside AI’s capability (spot a fraud!) none of these are prompting techniques. All transfer across tools and survive model upgrades. BCG study on using AI to do task “AI makes you 40% faster” — but it can also make you 19% slower! AI works in utter confidence even when it is wrong!! The Cognitive Debt Warning every time you hand AI a task, you are handing it a cognitive debt Part 5: Get in the Car