Aakash GuptaHow to Build Effective Product Loops in Claude Code | Tyler Folkman | Chief AI Officer, JobNimbus
CHAPTERS
- 0:00 – 2:27
Why “loops are the new prompts” for vibe PMing
Tyler and Aakash frame the core thesis: product work is shifting from single prompts to repeatable, self-improving loops. They set expectations that titles like PM/engineer/designer may matter less as AI raises individual leverage.
- •“Vibe PMing” as the PM analogue to vibe coding
- •Loops vs prompts: repeatable systems instead of one-off Q&A
- •AI makes work more people-centric, not less
- •Career implications: builders vs narrow function labels
- 2:27 – 4:48
Tyler’s daily loop-engineering workspace (Herder) and why terminal tools matter
Tyler demos Herder as a “pane of glass” for running multiple agents/loops in parallel. He explains why terminal-native workflows can outpace IDE-based tools as AI tooling evolves rapidly.
- •Herder orchestrates multiple agents and tracks status across tasks
- •Why terminal-based agents feel “native” for tool access and speed
- •Splitting panes, running editors + agents together
- •Community tools move fast because they’re built with AI
- 4:48 – 5:44
What makes something an agent (and a loop) instead of a prompt
They define “agentic” work as delegating tasks with verification so the model can make decisions autonomously. Tyler previews the “loop” concept as learning-driven improvement, not just execution.
- •Agent = autonomous execution + decision-making, not constant prompting
- •Verification is key to taking humans out of the loop safely
- •Loops close by learning from logs/outcomes
- •Self-improving flywheel vs static skill
- 5:44 – 8:46
Pre-flight checks and closing the loop: preparing for the podcast with AI
Tyler shows a real skill that performs a pre-flight checklist (safety, repo state, recording setup) before a live run. The example illustrates how loops enforce conditions for success and reduce “AI slop.”
- •Pre-flight checks: sensitivity, repo readiness, manual confirmations
- •Rehearsal vs live runs as a controllable workflow state
- •Closing the loop by reviewing logs and feeding learnings back
- •Loops create compounding improvement over time
- 8:46 – 12:13
Versioning and model drift: Git as rollback + adapting when “better” models regress
They discuss practical concerns: preserving intermediate versions and reverting changes. Tyler explains how Git enables safe iteration, and why newer models can sometimes perform worse on existing skills.
- •Git as the primary version-control mechanism for skills/loops
- •“Revert forward” vs reverting back: iterate toward the missing quality
- •Model drift: a better model can break a previously tuned skill
- •Importance of rollback + flexibility across model upgrades
- 12:13 – 14:43
The 4-part anatomy of a working loop: fetch → do work → gate → write artifact → learn
Tyler introduces a simple loop diagram and emphasizes the gate/validation step as the hardest and most important. They map this to real workflows: deterministic tests, LLM-as-judge where needed, and PR-based human review.
- •Fetch inputs (ideally the AI gathers its own)
- •Do work + validate via a clear “gate” (prefer deterministic)
- •Write artifacts (often via PR) after passing gates
- •Learning step turns a skill into a loop
- 14:43 – 17:24
From ideas to prototypes fast: generating multiple variants live
Tyler demonstrates how a product prototyping skill generates multiple prototype variants while they talk. The bigger point: AI drastically increases idea/prototype throughput—validation becomes the bottleneck.
- •AI generates multiple prototypes (minimal/full/creative variants) quickly
- •Prototype-first beats building the “one right thing” in production
- •Loop bottleneck shifts from creation to validation/gating
- •Rapid iteration enables more exploration with less build cost
- 17:24 – 19:51
Inside the skill library: domain teams author skills (UX prototyping as example)
They open the repo structure showing a `.claude` directory with commands/skills. Tyler explains how domain owners (like UX) codify their practices into reusable, semi-deterministic skills for consistent output.
- •Skills live in `.claude` and encode repeatable workflows
- •UX team owns prototyping skills; other domains own their tools
- •Skills translate principles into repeatable “canonical shapes”
- •System enables requesting ‘10 prototypes’ with consistent structure
- 19:51 – 22:22
Gating product ideas with customers: synthetic customers, transcripts, and scaling discovery
Tyler explains how they use call recordings/transcripts (warehoused) to create faster gates before real customer outreach. They aim to scale customer conversations across the whole org while maintaining research quality.
- •Use transcripts/recordings to synthesize insights and pre-filter ideas
- •Synthetic customer critique as a low-cost early gate (not final truth)
- •Goal: move 100 ideas → 5 strong candidates → real customer calls
- •Scaling customer contact: train everyone to avoid bad data/leading bias
- 22:22 – 28:33
AI-built onboarding flows and “docs written for AI” vs docs written for humans
Tyler shows an AI-generated onboarding roadmap that creates Linear boards and meeting plans, improving ramp time. They argue long docs should often be optimized for AI consumption, while human docs should be short, visual, and scannable.
- •Onboarding as a system: codify process first, then automate pieces
- •AI schedules meetings, creates Linear boards, and guides product learning
- •Docs for AI: high-context corpus the model uses to answer questions
- •Docs for humans: 1–3 pages, visual, succinct; avoid AI-length as ‘value’
- 28:33 – 42:47
Building a non-slop skill from scratch: human-authored constraints + compounding improvement
They write a simple “decide” skill manually, stressing that human-authored skills can outperform AI-authored ones. Tyler describes the “AI treadmill” risk and why you must selectively accept improvements to keep skills lean and useful.
- •Start by writing the first pass by hand (avoid outsourcing the thinking)
- •Define behavior: Socratic questions, contrarian checks, succinct output
- •Demonstration: the skill guides clarifying questions and decision framing
- •Closing the loop: ask how to improve the skill, but don’t auto-accept everything
- 42:47 – 49:27
Loops and hooks every PM should build: outreach, project health, post-session learning, safety
Tyler lists high-leverage PM loops (customer outreach, ticket/project synthesis) and explains hooks (startup, tool, and post-session) that add determinism and safety. Hooks can prevent dangerous actions (e.g., deleting files, exposing credentials) and enforce reflection.
- •PM loops: customer outreach targeting, project-management status synthesis
- •Post-session hook: analyze logs, identify friction, suggest improvements
- •Startup hooks: repo/branch/Git status context every session
- •Tool hooks: block dangerous commands (rm -rf), credential leakage, etc.
- 49:27 – 58:32
Quality loops and who should ship code: PM PRs, engineering safeguards, and code onslaught
They argue engineering must prioritize quality loops because shipping faster without quality improvement multiplies customer pain. Tyler gives nuance on PMs pushing code: it’s viable in well-guarded systems (tests, CI/CD) but must be agreed to avoid unplanned review load.
- •Quality loops: tests, standards checks, automated E2E as velocity rises
- •Customers feel absolute bug count, not defect rate
- •PM coding is situational: safer on decoupled front-end with strong guardrails
- •Process risks: review burden and “infinite code” generated by AI
- 58:32 – 1:08:53
Design loops and the rise of the product builder: triad convergence and career strategy
Tyler explains how design’s core loop is rapid prototyping and how AI enables non-designers to create acceptable UI using real design systems. They discuss the emerging “builder” role that blends PM/design/eng, hiring signals (hackathons), and how leaders create time to tinker.
- •Design loop: prototyping at speed; leveraging component libraries/design systems
- •“Vibe design” gradient: some work needs craft, some needs speed + adequacy
- •Builders create end-to-end value; specialists still matter for deep domains
- •Leadership adoption: carve IC time (meeting-light days) and lead by example
