Playbook
Practical guides for the people doing knowledge-intensive work — what to use AI for, what to keep human, and how to build a firm that compounds what it learns.
Practitioners who felt 20% faster with AI were measurably 19% slower. A framework for the only number that matters — net time saved after the verification and rework tax.
Three perimeters, three provider tiers, and a sensitivity ladder. A practitioner's guide to AI without disclosure risk — and what the structural answers actually look like.
Two questions, four zones, and a task-by-task map. A 60-second triage that decides whether AI lifts your work or sinks it — and where retrieval and memory shift the geometry.
Five things every senior practitioner needs to understand about how generative AI actually works in 2026 — why 95% of enterprise pilots fail, where the 40% rework problem comes from, and what the firms in the 5% are doing differently.
Why 'ground it in these documents' actually works, what good source-anchoring looks like in the output, and the retrieval patterns that separate a cited answer from a confident hallucination.
Spot-checking isn't a method. Five concrete, repeatable techniques — source-forcing, claim decomposition, adversarial re-prompting, citation checking, and the junior test — calibrated to what's actually at stake.
Prompt wording is the smallest lever. The input side — background, constraints, examples, voice samples, source documents — is where high-stakes output is won or lost. A pattern library for assembling it.
Six structural patterns that move output quality from 'meh' to 'usable' for high-stakes work — and where retrieval, voice memory, and the right tooling make most of them disappear.
Retrieval is only as good as what it retrieves. The information-architecture discipline — canonical sources, taxonomy, ownership, and what to deliberately exclude — that makes a knowledge base reliable instead of noisy.
MIT found 95% of enterprise AI pilots fail. The failure modes are knowable — and structural. How grounded retrieval, persistent voice memory, and hard confidentiality perimeters map onto each one.
Voice doesn't transfer in instructions — it transfers in artifacts. A four-stage cycle for teaching AI to write the way your firm sounds, and the three surfaces voice actually lives on.
The highest-stakes, most-searched PR-AI question, answered as a practical decision framework built on PRSA's 2025 ethics standard — material use, low-stakes assistance, and the gray zone between.
Reuters Institute projects a 43% drop in search referrals as AI answer-engines intercept the audience. A strategic reframing of earned media and message visibility for a world where ChatGPT, not Google, is the first stop.
The confidentiality map, applied to PR's realities — embargoed releases, client privilege, M&A and crisis work — and where a free chatbot quietly becomes a disclosure incident.
Three roles, one Tuesday, one boutique strategic communications firm. What changes when institutional knowledge stops walking out the door every night — for the account exec, the senior strategist, and the partner.
A concrete workflow for using AI on specific aims, broader-impacts, and RFP summarization — and the hard line on which proprietary sponsor language must never touch a public tool.
A practical template for the document a CIO, provost, or chief of staff actually needs — committee structure, data classification, procurement, and sunsetting — without the unenforceable jargon.
Presidential transitions, the enrollment cliff, and constant staff turnover all attack the same asset: what the institution knows about itself. Why persistent knowledge systems are the structural buffer.
Three offices, one Tuesday, one regional comprehensive. What changes when institutional knowledge stops walking out the door with every transition — for the director of sponsored programs, the dean of business, and the chief of staff.