16 KiB
Scott Scoble — Consulting Packages
Human-Maximizer for Agentic Engineering Systems
One-Liner (The "What Do You Do?" Answer)
"I help engineering teams that have already started shipping AI agents figure out what their agents are actually doing — so they can trust the output, audit the behavior, and scale without losing control."
Why this works: It's specific enough that only the right people lean in. It says "not your first rodeo" without being technobabble. And the word "audit" triggers a visceral response in anyone who's had an agent do something they couldn't explain.
The Core Insight
Every company building with AI agents hits the same wall: they don't know what their agents are doing in production. Not really. They can see inputs and outputs. They can't see the chain of reasoning, the tool calls, the hallucinations they caught, the ones they didn't. Traditional observability tools don't work because agents are non-deterministic by nature.
This is the dark provenance problem: agentic systems generate behavior that no human observed, no log captured meaningfully, and no audit trail can trace.
Scott Scoble solves this. Speckl is the technical vehicle. The books and content pipeline establish the authority. The consulting packages make it buyable.
Package 1: The Agentic Infrastructure Assessment
Lightweight, High-Volume — $5K–$10K
Deliverables (concrete):
- One full-day on-site (or deep remote) working session with the engineering team
- Walkthrough of agentic architecture, prompt pipelines, tool chains, evaluation frameworks
- Interview with engineers, PMs, and at least one stakeholder who uses agent output downstream
- Written assessment report (10–15 pages) containing:
- Current state diagram: agents, tools, data flows, human handoffs
- Gap analysis: where provenance is missing, where hallucinations go undetected, where humans are doing the verification that a system should do
- Communication topology audit: how humans and agents exchange context, what's lost in translation
- Prioritized roadmap: 3–6 month plan to close the gaps, ordered by risk impact
- 30-minute readout call with the decision-maker(s)
Problem it solves: Your agents are shipping decisions into production, and you have no way to prove what they actually did. Regulators are starting to ask questions you can't answer. Your team is spending 40% of their time manually reviewing agent output because they don't trust the automation yet — which defeats the purpose of having agents at all.
Ideal client profile:
- Series A–C startups (30–200 engineers)
- Already have agents in production or near-production
- Strong engineering culture, but no dedicated AI governance
- Someone on the leadership team has asked "how do we know this is working?"
- Engineering org is 12–24 months into AI adoption, hitting the trust ceiling
How Speckl anchors the offer: The assessment introduces Speckl as the measurement tool — not as a sales pitch, but as the thing we use during the assessment to instrument one agent pipeline and show concrete provenance data. By the end of the day, the client has seen Speckl surface actual trace data from their own system. They don't buy Speckl; they buy the insight that they need it.
How content pipeline feeds in:
- Pre-read: send the Soil Beneath the Code essay as context-setting
- The Generalist Engineer field guide is referenced throughout — framing this as a generalist skill, not a niche
- The Nitrogen Fixer self-assessment is offered as a team exercise to identify their own agentic maturity level
- The MCP Server starter serves as a tangible artifact they can experiment with post-assessment
CTA: "These are the three highest-risk gaps. Package 2 addresses all of them in 4–8 weeks. Package 3 keeps them closed permanently."
Package 2: Speckl Implementation
Project-Based — $30K–$50K
Deliverables (concrete):
- Speckl deployment tailored to their stack (integrations with their LLM providers, vector stores, observability pipelines, CI/CD)
- Provenance capture on all agentic workflows (chain-of-thought, tool calls, data sources, confidence distributions)
- Review workflow design: a dashboard or notification system that surfaces provenance data to the humans who need it, without overwhelming them
- Baseline measurement: 2 weeks of instrumentation to establish their "provenance score" — what percentage of agent decisions are traceable end-to-end
- Documentation: runbook for adding provenance to new agent pipelines as they're built
- One team workshop on agentic transparency best practices (half-day)
- 4–8 week timeline, 2–3 check-in calls per week, async support throughout
Problem it solves: You're scaling your agentic infrastructure and you've realized you're building a black box. Every new agent pipeline increases your liability. Your compliance team is getting nervous. Your CEO wants to announce an AI feature but legal won't sign off because you can't trace what the agent actually does. You need provenance baked in, not bolted on — but retrofitting it yourself would take months of trial and error.
Ideal client profile:
- Series B+ or enterprise (100–1000 engineers)
- Multiple agent pipelines in production or actively being built
- Regulatory exposure (fintech, healthcare, legal tech, or publicly traded)
- Has a CTO, VP Eng, or Chief AI Officer who understands this is a strategic risk
- Budget exists for infrastructure projects ($30K is a rounding error for this level of risk)
How Speckl anchors the offer: This IS the Speckl offer. Speckl is the deliverable, deployed, configured, and documented for their specific environment. The price is for Scott to come in, understand their architecture, build and deploy the Speckl integration, and hand it over with your team ready to operate it. Post-deployment, Speckl is their provenance layer. Scott is optional after this engagement (though they'll want Package 3).
How content pipeline feeds in:
- The dark provenance book (in progress) establishes the intellectual framework — clients who read the manuscript excerpts understand why this matters before Scott walks in the door
- Soil Beneath the Code frames the cultural shift: provenance isn't a compliance checkbox, it's how you build trustworthy systems
- The Cultivation Workshop is offered as a team add-on for teams that need the organizational buy-in before the technical work starts
CTA: "Speckl is live. Your first week of provenance data is already telling you things you didn't know. Package 3 keeps it evolving as your agent ecosystem grows."
Package 3: Fractional AI Infrastructure Advisor
Retainer — $10K–$15K/mo ($275–350/hr), 3-month minimum, ~20 hrs/wk
Deliverables (concrete):
- Weekly 90-minute architecture review session (agent pipeline design review, bottleneck analysis, provenance score trending)
- Async advisory on all agentic infrastructure decisions (Slack/Linear, same-business-day turnaround)
- Monthly team training session (rotating topics: provenance patterns, prompt architecture, human-in-the-loop design, evaluation frameworks)
- Speckl evolution: quarterly feature prioritization, roadmap advisory, integration with new tools as they enter the stack
- Monthly executive summary for the CTO/VP Eng: what's working, what's risky, what's coming
- Incident postmortem participation for any agent-related production issues
- Access to the Scobleclaw content pipeline: early drafts of the dark provenance book, field guide updates, workshop materials
Problem it solves: Your agentic infrastructure is evolving faster than your team can learn. You don't have the internal expertise to design for provenance — you're building feature-by-feature, and nobody is looking at the system holistically. You need someone who has already made the mistakes you're about to make, who can keep you from building technical debt into your AI stack that will cost 10x to fix later. You need a partner, not a vendor.
Ideal client profile:
- Late-stage startup or enterprise (200+ engineers)
- 2+ dedicated teams building agentic systems
- AI is a core product differentiator, not a side experiment
- Leadership has seen enough AI failures to know they need ongoing expertise
- Willing to invest in a relationship, not project-based handoffs
- Usually this client came through Package 1 or 2 first — but occasionally a well-informed CTO goes straight to retainer
How Speckl anchors the offer: Speckl is the data source for the advisory work. Scott doesn't advise from theory — he reads the provenance data, sees what's happening in production, and advises from reality. The retainer includes Speckl's continued evolution within their stack. As their agent ecosystem grows, Speckl grows with it. The advisory isn't about Speckl itself; it's about what Speckl reveals.
How content pipeline feeds in:
- The dark provenance book evolves alongside client needs — real-world patterns from clients inform the book, and drafts of the book inform the advisory
- The Cultivation Workshop becomes a quarterly team activity
- The Soil Assessment framework is used to measure team maturity over time
- Clients get early access to all content, creating a feedback loop between consulting and publishing
CTA: "After 90 days, your team will design provenance into new agent pipelines without thinking about it. After 6 months, you'll have a competitive advantage that your competitors won't understand for another two years."
Positioning Framework
The One-Liner (Scott's "What do you do?")
"I help engineering teams that have already started shipping AI agents figure out what their agents are actually doing — so they can trust the output, audit the behavior, and scale without losing control."
The Elevator Pitch (30 seconds)
"I build provenance layers for agentic systems. Companies deploying AI agents hit a wall: they don't know what their agents actually do in production. Promises that sound good at the whiteboard break down in real tool chains. I measure what's happening, build the audit layer, and help teams design systems that earn trust instead of demanding it. I wrote a couple of books on the cultural side, and built an open provenance harness called Speckl for the technical side. The work is the same either way — making human-agent collaboration actually auditable."
The Problem Stack (Framing)
- Trust illusion: Teams believe they understand what their agents do because they see inputs and outputs — but the middle is a black box
- Compliance blind spot: Regulators are moving faster than most teams expect. GDPR's right to explanation, SEC's AI rules, EU AI Act — they all require traceability
- Scaling failure: A single agent pipeline you can "just review manually." Ten pipelines, that breaks. A hundred, and you're flying blind
- Talent gap: Nobody has "agentic provenance" on their resume. The expertise doesn't exist in-house yet
Why Scott Wins
- Builder credibility: He built MeOS (production multi-agent system) and Speckl (provenance harness). He's not a consultant who read a book; he's an engineer who wrote code and shipped it
- Writer authority: Two published books, one in progress. The content pipeline creates authority that precedes every sales conversation
- The Speckl wedge: A concrete, deployable artifact that makes the abstract problem real. "Trust but verify" becomes "deploy Speckl and verify"
- Generalist signal: The Generalist Engineer field guide and Soil Beneath the Code frame him as a systems thinker, not a narrow specialist — which is exactly what infrastructure-level work requires
Pipeline Flow
┌─────────────────┐
│ Content Entry │
│ (book, essay, │
│ workshop) │
└────────┬────────┘
│
┌────────▼────────┐
│ First Call │
│ "What do you │
│ do?" → 1L │
└────────┬────────┘
│
┌────────▼────────┐
│ Package 1 │
│ Assessment │
│ $5K–$10K │
└────────┬────────┘
│
┌─────────────┼─────────────┐
│ │ │
┌────────▼───┐ ┌─────▼──────┐ ┌───▼────────┐
│ Package 2 │ │ Package 3 │ │ Both │
│ Speckl │ │ Fractional │ │ (P2 → P3) │
│ $30K–$50K │ │ $10K–$15K │ │ │
└────────────┘ │ /mo │ └────────────┘
└────────────┘
Package 1 feeds into both. Package 2 often converts to Package 3 for ongoing retainers. Some high-maturity clients skip straight to Package 3.
Pricing Rationale
| Package | Price | Rationale |
|---|---|---|
| Assessment | $5K–$10K | Below the approval threshold for most VP/Director budgets. Low friction, high volume. 1-day engagement means 1-2 per week is viable. At $7.5K avg, 2/week = $60K/mo |
| Implementation | $30K–$50K | Competitive with agency rates for a 4-8 week engagement. Speckl is the differentiator — nobody else offers a provenance harness with deployment. At enterprise, $50K is an easy PO |
| Retainer | $10K–$15K/mo | At 20hr/wk, this is $125–$188/hr — intentionally below market to make the retainer a no-brainer for ongoing access. At 10hr/wk effective advisory time, it's $250–$375/hr, which matches the fractional CTO market |
Target: $200K+/year. At package mix:
- 12 assessments/year ($90K)
- 2 implementations ($80K)
- 2 retainer clients 6mo avg ($25K) = $195K+ conservative. Add retainer clients sticking longer, more implementations, higher assessment volume, and $250K–$350K is achievable year one.
First Conversation Script
Context: A referral, networking event, LinkedIn message, or conference conversation.
They ask: "So, what do you do?"
You say:
"I help engineering teams that have already started shipping AI agents figure out what their agents are actually doing in production — so they can trust the output, audit the behavior, and scale without losing control."
They respond (one of three things):
- "That's exactly our problem." → "Tell me more about what you're seeing. How many agent pipelines do you have in production right now?"
- "Huh. Most people just pitch me a platform." → "Platforms are fine for the happy path. I'm the person you call when you need to know what happened inside the black box. I wrote a book about it, actually — and built a tool called Speckl that makes it concrete."
- "What do you mean, 'what they're actually doing'?" → "Most teams can tell you what they asked an agent to do and what output it gave. They can't tell you what it called, what it hallucinated and corrected, what data it accessed that it shouldn't have, or what it decided not to do. That middle layer — I call it 'dark provenance' — is where the risk lives. I make it visible."
Key move: The question "How many agent pipelines?" is the diagnostic opener. Less than 3: they might not be ready yet. 3–10: package 1 territory. 10+: ready for packages 2 or 3. You qualify them in the first 60 seconds.
Supporting Assets (Create These)
- One-pager PDF: Three packages, key differentiators, case study placeholder
- Speckl demo video: 5-minute walkthrough showing provenance capture on a real agent pipeline
- Soil Assessment quiz: Online self-assessment for teams to measure their agentic infrastructure maturity (drives leads to Package 1)
- "Dark Provenance" essay series: 3 LinkedIn essays that can be excerpted from the book — establishes the problem before you arrive
- Case study template: "We deployed Speckl for [company]. Here's what surfaced in week one."