AI Literacy for the Modern Workforce

Problem

Work­force AI adop­tion has out­run work­force AI judg­ment. The World Eco­nom­ic Forum’s 2025 Fu­ture of Jobs Re­port finds 63 per­cent of em­ploy­ers nam­ing skill gaps as the pri­ma­ry bar­ri­er to AI-dri­ven trans­for­ma­tion, while pro­duc­tiv­i­ty re­search doc­u­ments an 81 per­cent me­di­an task-time re­duc­tion in­side the AI con­ver­sa­tion against 14–56 per­cent across the full work cycle — a dis­tance that mea­sures miss­ing com­pe­ten­cy, not miss­ing ac­cess. The train­ing on offer most­ly teach­es tool fa­mil­iar­i­ty and mea­sures com­ple­tion. It does not build the judg­ment to de­cide what to del­e­gate to AI, how to spec­i­fy it, whether to trust what comes back, and how to stay ac­count­able for the re­sult — and it rarely pro­duces ev­i­dence of be­hav­ior change that would jus­ti­fy con­tin­ued in­vest­ment.

Constraints

The pro­gram was built solo, along­side a full-time role, in rough­ly eight cal­en­dar weeks. Three self-im­posed con­straints shaped it more than the time­line. Every sec­tion had to trace back­ward to a doc­u­ment­ed gap in the re­search cor­pus and for­ward to a mea­sur­able as­sess­ment — any­thing that could not be jus­ti­fied by ev­i­dence and mea­sured by an ob­serv­able be­hav­ior was cut. The eval­u­a­tion frame­work had to be de­signed be­fore the mod­ules ex­ist­ed, so suc­cess was de­fined be­fore there was any­thing to grade. And no off-the-shelf au­thor­ing tool: in­ter­ac­tion de­sign and data cap­ture could not in­her­it a ven­dor’s con­straints, which meant build­ing the plat­form it­self.

Approach

Four mod­ules — Con­text, Ev­i­dence, Mech­a­nism, Ap­pli­ca­tion — take a learn­er from the busi­ness case, through doc­u­ment­ed usage ev­i­dence, into how lan­guage mod­els ac­tu­al­ly gen­er­ate text, and fi­nal­ly into ap­plied prac­tice. An­throp­ic’s 4D com­pe­ten­cy frame­work (Del­e­ga­tion, De­scrip­tion, Dis­cern­ment, Dili­gence) pro­vides the spine that con­nects 17 per­for­mance ob­jec­tives to ob­serv­able work­place be­hav­iors across 37 sec­tions and 12 prac­tice ac­tiv­i­ties, in­clud­ing a to­k­eniz­er play­ground, task-level adop­tion dash­boards, and a multi-step AI in­ter­ac­tion sand­box.

Eval­u­a­tion is built on the four Kirk­patrick lev­els: a re­ac­tion in­stru­ment in which every item dri­ves a spe­cif­ic pro­gram change, a sce­nario-based par­al­lel-form pre/post as­sess­ment car­ry­ing real val­i­da­tion data, a 30/60/90-day man­ag­er ev­i­dence re­view scored on the 4D frame­work, and a KPI and ROI model that con­nects be­hav­ior change to or­ga­ni­za­tion­al out­comes. The plat­form cap­tures progress, knowl­edge-check re­spons­es, and time-on-task through an xAPI-aligned event tax­on­o­my sur­faced in a built-in an­a­lyt­ics dash­board — the re­port­ing a learn­ing-and-de­vel­op­ment di­rec­tor needs, gen­er­at­ed by the ar­chi­tec­ture rather than bolt­ed on.

The build it­self ran on a spec­i­fi­ca­tion–ver­i­fi­ca­tion loop with AI de­vel­op­ment tools: de­tailed build spec­i­fi­ca­tions, AI-gen­er­at­ed im­ple­men­ta­tion, ex­pert re­view against the spec­i­fi­ca­tion and the source re­search. The method is the cur­ricu­lum — the same del­e­ga­tion-and-dis­cern­ment cycle the pro­gram teach­es, ap­plied to pro­duc­ing it.

Outcome

The pro­gram is live at ai-literacy.ritchot.me as a v1.0 port­fo­lio edi­tion. Total de­vel­op­ment ran to rough­ly 150–160 hours against an in­dus­try base­line near 735 hours for com­pa­ra­ble in­ter­ac­tiv­i­ty (Chap­man’s bench­mark for sim­u­la­tion-grade con­tent) — a 4.6–4.9× com­pres­sion con­sis­tent with doc­u­ment­ed ranges for AI-aug­ment­ed knowl­edge work, achieved with­out sac­ri­fic­ing the ev­i­dence trail. The com­plete pro­fes­sion­al record is pub­lished with the course: the needs analy­sis, the eval­u­a­tion frame­work, and the de­sign and project-man­age­ment records — time­line, re­spon­si­bil­i­ty ma­trix, bud­get, stake­hold­er com­mu­ni­ca­tions, and QA — as in­ter­ac­tive walk­throughs and down­load­able doc­u­ments, re­fram­ing the solo build as the five-per­son or­ga­ni­za­tion­al de­ploy­ment it mod­els.

Stack and links

React, Type­Script, Vite, Tail­wind CSS, and Recharts; gpt-to­k­eniz­er dri­ves the to­k­eniz­er play­ground. No back­end — state lives client-side, an­a­lyt­ics events ex­port as JSON/xAPI — de­ployed sta­t­ic on Cloud­flare Pages. Code is MIT-li­censed; in­struc­tion­al con­tent is CC BY-NC-SA 4.0.

• The live program
• Source on GitHub
• Needs analysis · evaluation framework · build records
• Com­pan­ion essay: I built an AI Literacy course