SOURCE

Why Your AI Offer Isn't Selling, and How to Fix That

Properties1
tagsai-agency, offer-positioning, storytelling, ai-native, claude-code, verification, nate-herk, nate-b-jones
AUTHOR
Nate B. Jones (guest); Nate Herk (host)
PUBLISHED
2026-07-20
ORIGINAL
view original

6 min read

Why Your AI Offer Isn’t Selling, and How to Fix That

Guest: Nate B. Jones · Host: Nate Herk | Published: 2026-07-20 | Source: transcript (55:40, 39.9K views at ingest)

Summary

Nate Herk interviews strategy commentator Nate B. Jones for AI-agency builders who can scope and deliver but stall on selling. Jones’s thesis: every company he talks to — 10-person shops through orgs with hundreds of thousands of employees — already wants AI. If the offer isn’t closing, the problem is storytelling, not demand. AI itself is not the product. Sell the transformation (for an HVAC shop: “I pick up the phone” / 24/7 dispatch, not “we implement AI”). A legal-firm anecdote: vendors listing security certifications all lost to the one who plunked a server on the table and said “your data lives here and it won’t leave.”

He frames builders as information-arbitrage operators for the next ~5 years — the electricity-into-the-factory era. Handing employees buckets of Claude credits is “giving a 480-volt sparking live line to a factory worker and saying figure it out.” Hyperscalers cannot do last-mile org design (that’s why they hire thousands of forward-deployed engineers); agencies and operators can. Tactic: pick a specific demographic (his example: Seattle HVAC — same niche Caleb Ulku uses) and go offline, because those buyers aren’t competing with agents for attention.

On the operator side: AI-native means reaching for AI first, second, and last — a “mech suit.” Vibe coding is cheap; the missing piece is an intent-verification loop (correct, sustainable, secure) — same process layer as AI-Assisted Development. Token usage is the tell: Jones hits ~100M tokens/day personally (sometimes 800M); in hundreds-person teams, power users burn ~2B/day while most sit at 2–5M and “aren’t doing anything productive.” Scale usage with a catalyst project that is big, near-term, and a little scary, plus an auto-router (95% open models, frontier when needed). AI-native teams are small (Midjourney cited at $200M run-rate / 40 employees). Transformation is team-leader-led (SDM, ~10–12 people), not org-wide. IBM’s 2000-CEO survey (Herk: 76% want leaders tech-fluent) — Jones agrees; non-tech C-suite (marketing, sales, finance, legal, HR) must get technical, and the CEO must personally use Codex / Claude Code or nobody else will.

Hot takes: jobs aren’t going away because jobs ≠ tasks; most of AI’s quality-of-life impact is still outside work; the 50-year story is open-source models + AI-optimistic countries (UAE), not hyperscaler bench scores. Taste is the differentiator as AI content commodifies (Linear’s look as the example of taste that then got copied). Commercial layer: Herk funnels a free Skool community + ClickUp affiliate; Jones funnels Substack/Slack + OpenBrain GitHub.

Key Claims

  • If an AI offer isn’t selling, the builder has a storytelling problem, not a demand problem — buyers at every scale already want AI. (Jones, creator-opinion)
  • Don’t sell AI. Sell the outcome/transformation (HVAC: pick up the phone 24/7). (Jones)
  • Verification > generation: vibe coding communicates intent; leaders need an intent-verification loop that the result is correct, sustainable, and secure. 2027 investment theme. (Jones; aligns with AI-Assisted Development)
  • AI-native = reach for AI first, second, and last; treat it as a mech suit. (Jones)
  • Tool selection heuristic: does this release take weight off your shoulders? If not, skip it. (Jones)
  • Information-arbitrage window ~5 years — factories being rewired around electricity, not overnight. Claude-credit buckets without design = live 480V line. (analogy)
  • 100M tokens/day is a “manageable bar”; team laggards at 2–5M aren’t using AI productively. Fix with a near-term scary catalyst project + auto-router (95% open models). (self-reported)
  • AI-native teams are small; transformation is SDM-level team-leader-led, not org-wide. C-suite outside product/eng must get technical; CEO must personally use Codex/Claude Code. (Jones; Herk cites IBM 76%)
  • In a Karpathy-style world of 99% agent attention / 1% human attention, human relationships are the undebeasable currency. Pick a specific offline demographic. (attributed + Jones)
  • Jobs survive because jobs ≠ tasks; the long-run AI story is open-source + AI-optimistic countries, not closed hyperscaler labs. (Jones, hot take)

Notable quotes

“If you are having trouble selling, the issue is that you are having trouble storytelling. You shouldn’t be selling AI. AI is not valuable in and of itself. Sell that story. Sell the transformation.”

Why citable: The episode’s title claim in one block — the positioning rule for AI-service offers, and a direct tension with selling “AI presence” as the product.

“Handing out Claude credits to your employees in buckets is the same thing as giving a 480 volt sparking live line to a factory worker and saying figure it out.”

Why citable: The electricity-factory analogy that distinguishes tool access from org design — the actual last-mile job agencies/operators sell.

“The buck stops with you. You are the AI change you want to see in the world. If you are not using Codex, if you are not using Claude Code, don’t expect other people to.”

Why citable: Top-down enablement as a hard constraint, not a culture poster — useful for any client-side AI rollout pitch.

Connections

Entities mentioned: Claude, Claude Code, ChatGPT, Caleb Ulku (adjacent: HVAC niche + agency selling) · NEW Nate Herk, NEW nate-b-jones · OpenAI Codex, Anthropic, Midjourney, Shopify, IBM, Linear, ClickUp (sponsor), Alex Hormozi (minor) Concepts referenced: AI-Assisted Development (verification loop, vibe coding, Claude Code) · NEW Agent Orchestration (token budgets, auto-router, catalyst projects, small AI-native teams) · Core30 AI Tool (agency AI-tooling analog)

Contradictions / Tensions

  • vs. Caleb Ulku on what you sell: Caleb’s GEO thread tells agencies to sell “AI presence” (TRAIN Any AI To Recommend You (This Study Proves How)); Jones says never sell AI, sell the operational outcome. Same buyer (local/SMB, even the same HVAC example), opposite packaging. Record both — they can stack (outcome story on the outside, AI-presence work on the inside) but they are not the same offer.
  • Agrees with AI-Assisted Development / Paste This Into Claude, It'll Make You Build 10x Faster (6 Claude Code power phrases) on verification, spec/intent, and “don’t automate blindly”; Jones adds the sales and org-change layers Marchese doesn’t cover.
  • Jobs-not-screwed is a hot take against the replacement narrative Herk reports from commencement-speech backlash; Jones attributes layoffs more to ZIRP overhiring + uncreative boards than to task automation (cites Jensen Huang).
  • Commercial bias: dual funnel (Herk Skool + ClickUp; Jones Substack/Slack). Token-usage figures and Midjourney headcount/ARR are asserted, not sourced. IBM 76% is Herk’s recollection of a “2000 CEO survey.”

Notes

Adjacent ingest (not core local-SEO) but high-value for the practitioner/agency-ops thread: how to package AI work for SMBs. HVAC-in-Seattle targeting is the same niche pattern Caleb uses for local SEO. Backfill candidates: Nate B. Jones’s “no monsters in the forest” X essay; IBM 2026 CEO survey for the 76% figure; Midjourney 40-employee / $200M claim. NEW entities to create later: nate-herk, nate-b-jones. NEW concept if this cluster grows: agent-orchestration.