Valuestream

Rick Pollick
  • 0.0 (0)
  • Technology
  • Updated Monthly

A podcast from Rick Pollick on how modern companies actually turn strategy into shipped software. Every episode walks one real story through three segments — intake, flow, outcome — covering operating models, platform engineering, and agentic-AI patterns that move value from the roadmap to production. 25 to 35 minutes, monthly. No vendor pitches. Focus is the operating model, not the tool stack. Companion essays at rickpollick.com/blog.

Episodes

  1. 22h ago

    Ep. 5 — Context Is the Job: Reliable AI Agents Are a Context Problem, Not a Prompt Problem

    A team rewrote the system prompt nineteen times. The agent kept giving customers the same confidently wrong answer. Then someone logged what the model actually saw: a policy doc from 2023, nine hundred lines of raw tool output, and the correct policy buried in the middle of forty thousand tokens where the model never looked. The prompt was fine. The context was garbage. The frameworkThe context budget. The four failure modes. The operating model. Context is the job. Inside this episode Intake — Prompt engineering was the tutorial; context engineering is the job. A 2026 survey traced 57% of enterprise agent reliability failures to missing or inconsistent context, not the model. The fix is a delivery discipline: govern what the agent can see.Flow — The context budget (attention degrades before the window fills), the four failure modes (retrieval, tool-output flooding, memory that never forgets, history that never compacts), and the operating model: version control, evals, and observability, with a named owner.Outcome — A support agent that was confidently wrong 22% of the time drops to about 4% on the same model, by fixing what it could see: context cut from ~38k tokens to ~12k, retrieval quality 60% to the low 90s, and mean time to understand incidents measured in minutes.Also insideWhy more context is not more capability (the "lost in the middle" effect); the retrieval-reasoning split that cut human-in-the-loop rates 30-45% in BCG deployments; and why the best context change you make this quarter is probably something you remove. Links Full transcript & framework: rickpollick.com/blog/valuestream-episode-5-context-is-the-jobShow home: rickpollick.com/podcastNext episodeProgressive delivery in the AI era: canary releases, feature flags, and automated rollback wired to an error budget. Speed without a kill switch is just faster chaos. Subscribe in your podcast app of choice. New episodes monthly. Companion essay on the blog with each ship. This is Valuestream. I'm Rick Pollick. Context is the job.

  2. Aug 10

    Ep. 4 — The Demo Isn't the Deliverable: Crossing the Production Gap From Pilot to Production

    A pilot "killed it" in the demo: eight seconds to read a customer email, pull the account, and draft a refund. The room clapped. Five months later it hasn't touched a single real customer. It didn't fail. It stalled in the production gap. The frameworkVerification. Evaluation. Operability. Ownership. Cross the gap one layer at a time. The demo isn't the deliverable. Inside this episode Intake — The reason your pilot isn't in production has almost nothing to do with the model. MIT finds ~95% of enterprise AI pilots show no P&L impact. The gap is the delivery and operating work between a demo that works once and a system that survives reality.Flow — The four layers of the gap, walked concretely: verification (review capacity), evaluation (evals as the new acceptance criteria), operability (the trace, rollback, error budgets), and ownership (fund a product, not a project).Outcome — A composite program stops starting pilots, picks one, and runs it across the gap: reviewer utilization 96% down to a sane queue, zero evals to 300+, five months stuck to shipped in seven weeks, then the next pilots cross in a fraction of the time.Also insideWhy generation got cheap but verification didn't (AI pull requests run 2.6x larger and merge at ~33% vs ~84%); the queue math that makes review detonate near 95% utilization; and why internal builds reach production about a third as often as bought-and-partnered ones. Links Full transcript & framework: rickpollick.com/blog/valuestream-episode-4-the-demo-isnt-the-deliverableShow home: rickpollick.com/podcastNext episodeThe runway itself: platform as a product, the paved road from demo to production that turns crossing the gap from a heroic act into a Tuesday. Subscribe in your podcast app of choice. New episodes monthly. Companion essay on the blog with each ship. This is Valuestream. I'm Rick Pollick. The demo isn't the deliverable.

  3. Jul 10

    Ep. 3 — The Agent Acts, You Answer: Governing the Agents You've Already Deployed

    An AI agent deletes a payments safeguard that existed because of a disaster. The pull request is green. The reviewer is junior. Nobody ever decided the agent was allowed to touch that service. It just could. The agent acted. Who answered? The frameworkScope the authority. Keep the trace. Right-size the human checkpoints to blast radius. Give the agent memory. Red-team the whole thing. The agent acts, you answer. Inside this episode Intake — Once agents act and chain decisions across your systems, "is this model accurate" stops being enough. The new question is what an agent is allowed to do, and who's accountable when it acts.Flow — The five-move control model: capability envelopes enforced at the infrastructure level, the trace, tiered human checkpoints, ADRs as the agent's memory of why, and governance red teams.Outcome — A model program runs the five moves in a quarter: ADR coverage 35% to 78%, agent-authored drift from ~62% to single digits, and the trace catches a near-miss before it ships.Also insideGartner expects agents in 40% of enterprise apps by the end of 2026 and 150,000+ per Fortune 500 by 2028; non-human identities already outnumber humans roughly 80 to 1; and why one-size-fits-all agent governance actually backfires. Links Full transcript & framework: rickpollick.com/blog/valuestream-episode-3-the-agent-acts-you-answerShow home: rickpollick.com/podcastNext episodeBoundaries got us here. Next time it's capability: giving the agent a brain, the context, memory, and priorities that turn a confidently-wrong assistant into something that actually knows your week. Subscribe in your podcast app of choice. New episodes monthly. Companion essay on the blog with each ship. This is Valuestream. I'm Rick Pollick. The agent acts, you answer.

  4. Jun 12

    Ep. 2 — Trust the Number That Hurts: Metrics That Lie in the Agentic Era

    Two leaders. Two trusted numbers. Two metrics that looked great and lied. Episode 2 of Valuestream walks the delivery metrics that were built for a pre-agentic world and quietly stopped working the moment agents started authoring code and dragging dependencies into your programs. The thesisThe cost of a metric that lies isn't that it's wrong. It's that it's confident. It walks into the board meeting, flashes green, and buys you another quarter of believing motion is progress. Inside this episode Intake — Why DORA's four metrics and program-level critical path are both failing in the same way: the assumption that the thing producing the work is human has changed, and the math hasn't.Flow — The five-metric replacement layer for engineering (outcome latency, reviewer load, rework ratio, time to restore, production confidence) plus dependency density for programs (edges per node, with four edge types: code, data, process, people).Outcome — A composite case at 200 deploys/week reporting elite on every DORA number. Segmenting failure rate by author type showed agent-authored failure at 3x the human rate. Rework ratio hit 26%. The payments cluster had a dependency density of 3.4. We collapsed services, decoupled the integration, and cut the highest-edge scope. Density dropped to 1.9. Rework dropped to 9%. Deploy count went down on purpose. The program shipped on a date we could actually predict.The migration planDon't swap your dashboards overnight. Quarter one: instrument the new numbers next to the old ones without acting. Quarter two: add reviewer load and the confidence score, segment failure rate by author type. Quarter three: demote deploy frequency and critical path from headline to context. This week's prescriptionTwo spreadsheets. Pick one program. Segment last month's change failure rate by author type (human, agent assisted, agent authored). Score your three highest-risk clusters for dependency density. Look at the agent-authored failure number sitting next to the blended one you've been reporting. Trust the number that hurts. Links Full transcript and frameworks: rickpollick.com/blog/valuestream-episode-2-trust-the-number-that-hurtsEpisode 1 (See It, Own It, Move It): rickpollick.com/blog/valuestream-episode-1-see-it-own-it-move-itShow home: rickpollick.com/podcastNext episodeThe governance version of this conversation. Once an agent authors a third of your changes and chains decisions across your systems, the old question "is this model accurate" stops being enough. The new question is "what is this agent allowed to do, and who's accountable when it acts." Subscribe in your podcast app of choice. Companion essay on the blog with each ship. This is Valuestream. I'm Rick Pollick. Trust the number that hurts.

  5. May 25

    Ep. 1 — See It, Own It, Move It: Where Value Actually Flows

    The launch episode of Valuestream. Nine senior people in a room, forty-five minutes blocked, one decision on the table — and nobody makes it. This is the quietest failure mode in modern delivery, and the three-word fix you can ship in a Google Sheet on Monday. The frameworkSee it. Own it. Move it. Visibility is the intervention. One name per decision. Every decision gets a date — not the ship date, the decide date. Inside this episode Intake — three quiet failure modes (handoff theater, update chasing, decorative dashboards) and what decision latency actually costs you ($50K per incident; $3,750 per employee per year in lost productivity per McKinsey).Flow — wiring Owner-Decision-Date into a product-team operating model, an internal developer platform, and an agentic-AI strategy that survives 2027 (Gartner: 40% of agentic AI projects will be cancelled by end of 2027 — the cause is organizational, not technical).Outcome — a 60% drop in decision latency in two weeks (51 days → 19), a platform team that took mean-time-to-production from 14 weeks to 3, and an agent metric change that doubled effective throughput.Also insideWhy 80% of large engineering orgs will have a dedicated platform team by end of 2026 (Gartner) and what most are getting wrong. Trust calibration, hybrid workflows, and feedback loops for agentic AI. Why 97% of orgs have hit an AI security incident and what governance actually needs to cover now. The five-step re-baselining conversation that turns a slipping program back into a managed one. Links Full transcript & framework: rickpollick.com/blog/valuestream-episode-1-see-it-own-it-move-itHeatmap template + Product Camp Pittsburgh deck: rickpollick.com/blog/not-my-problem-product-camp-pittsburgh-2026Show home: rickpollick.com/valuestreamNext episodeHow to introduce decision visibility when your culture punishes visibility — every CFO has nodded at the heatmap and then said "I can't show this to my CEO." Subscribe in your podcast app of choice. New episodes every other week. Companion essay on the blog with each ship. This is Valuestream. I'm Rick Pollick. See it, own it, move it.

  • 5 Episodes

About

A podcast from Rick Pollick on how modern companies actually turn strategy into shipped software. Every episode walks one real story through three segments — intake, flow, outcome — covering operating models, platform engineering, and agentic-AI patterns that move value from the roadmap to production. 25 to 35 minutes, monthly. No vendor pitches. Focus is the operating model, not the tool stack. Companion essays at rickpollick.com/blog.

Information

  • Creator
    Rick Pollick
  • Years Active
    2026
  • Episodes
    5
  • Rating
    Clean
  • Copyright
    © Rick Pollick
  • Show Website
    Valuestream