The One Big Beautiful Bill Act rewrote 174 expensing rules in July 2025. Section 41 four-part test enforcement has tightened in parallel. And every CPA with R&D credit clients is relearning the game in real time, while AI/ML companies - the fastest-growing segment of qualified claimants - flood the market with software development costs that may or may not qualify.
This Playbook is for the practitioner who needs to get this right. Not a general overview. A working manual for qualifying, computing, and defending 41 claims for AI/ML clients in the post-OBBBA environment, with AI used as both subject (what counts as qualified research for AI development) and tool (how to surface QREs and draft audit-defensible technical narratives at scale).
What's inside: Four-part test checklists for AI/ML projects - model architecture, training pipelines, fine-tuning (qualifying); prompt tweaking, off-the-shelf deployment (not qualifying)27 sample QRE narratives covering common AI development patterns, ready to adaptPost-OBBBA 174 / 41 computation walkthroughs with worked examplesAI-assisted documentation workflows that hold up under IRS examAudit defense templates: substantiation memos, time-tracking frameworks, contemporaneous documentation protocolsEngagement letter language and scoping templates for R&D credit workRed flags: positions that look qualifying but won't survive examinationWritten for: CPAs running R&D credit engagements, R&D specialty firm consultants, in-house tax directors at AI/ML companies, tax attorneys advising on 41 positions, and tax controversy attorneys defending 41 credits under examination.