NameIntel

Day 18 of 30 — the Autopilot arc opens

NameIntel is a brand-naming intelligence service built for the buyer the web is acquiring: the AI agent. It is live today at nameintel.io, and it is live as an MCP server — score_name, check_domain, check_trademark, check_social, check_geo — that any MCP-capable client can call. The same endpoints accept x402 micropayments in USDC on Base or Optimism mainnet, so an agent can pay per query without a human credit card in the loop. Co-founded with Jim Hasty — the same partnership the run formalized as a GP seat inside CogleGroup on Day 9.

A note on the schedule. The original calendar had NameIntel on Day 16 (Saturday) and Planwright on Day 17 (Sunday). The weekend slipped — the x402 mainnet path took an extra two days of facilitator-integration debugging, and shipping the launch with a testnet-only disclosure would have undercut the production-form argument the post is making. Both builds compressed into this opening. Day 18 lands now. Day 19 — Planwright — lands tomorrow. The run is public and the discipline is to name slips when they happen.

Day 15 was the halftime — a Velocity Process cohort site build for Swole Labor Services in Fort Myers, and the marker that the run was 15 days in and 15 days to go. Day 18 turns the run into its second-half thesis: the Autopilot economy. The frame is the one Sequoia named in Services: The New Software — sell the work, not the tool — and NameIntel is the cleanest expression of it in the run: a product whose primary buyer is not a human at a keyboard, but a software agent that has a budget and a job to do.

What it does

NameIntel takes any candidate brand name and returns five sub-scores plus a composite, with a one-line verdict. The dimensions are deliberately the five surfaces a founder used to check by hand.

Domain (25%). Availability and live pricing across the six TLDs that matter for a launchable brand — .com, .io, .ai, .app, .dev, .co — with caller-supplied TLD overrides. The check runs against AWS Route53 Domains for live registrar pricing where Route53 covers the TLD, and falls back to RDAP for TLDs Route53 doesn’t quote. The result names the source per TLD so the caller can audit it.

Trademark (25%). USPTO TESS exact-match and similar-mark search, with Nice-class extraction for the marks that match. The output is none / low / medium / high risk plus the underlying mark count and classes. This is the dimension that prevents a founder from getting six months into a brand they can’t legally use.

Social handles (15%). Availability check across eleven platforms — Instagram, Facebook, TikTok, YouTube, GitHub, LinkedIn, X, Threads, Reddit, Pinterest, Medium, Snapchat. Platforms that block unauthenticated probes return an explicit unknown state, not a guessed answer. The honesty in the unknown-state is load-bearing for an agent caller: a hallucinated “available” wastes a registration attempt.

SEO (15%). Search-volume tier, competitor density, and keyword relevance — derived from a deterministic prompt and a cached corpus rather than a paid search-volume API on the free path. A name like “Apple” scores low here; an invented compound like “NameIntel” scores high. The output is qualitative (high / medium / low) plus a one-line rationale.

GEO — generative-engine findability (20%). This is the dimension nobody else in the naming category surfaces. Claude Haiku on AWS Bedrock scores entity collision, semantic distinctiveness, corpus presence, and phonetic clarity, with a deterministic prompt that returns the same answer for the same input within a 30-day window. The score is the answer to “if someone asks an LLM for ‘best X,’ does this brand land in the answer?” — and it is the dimension the rest of the category is still pretending doesn’t exist.

The composite is a weighted sum that returns a 0–100 score and one of five verdicts: Strong — pursue · Solid — minor concerns · Moderate — investigate concerns · Weak — significant issues · Poor — strongly reconsider. A founder gets the answer in a single call. An agent gets it in a single MCP tool invocation, with a signed payment proof attached.

The two-surface launch

NameIntel ships today on two surfaces, and the structural argument of the launch is that they are the same product:

The website surface — nameintel.io. A clean console: enter a name, get the five-dimension breakdown and the composite, see live domain pricing per TLD, and inspect the trademark and social-handle detail. This is the surface for a human founder who is shortlisting names the old way.

The MCP / x402 surface — mcp.nameintel.io. The same five scoring functions, exposed as MCP tools. Point Claude Desktop, Cursor, Cline, or any agent runtime at the server and the tools light up — score_name("Vibestack") returns the same composite the website does, in the same JSON shape, in one call. The endpoints are x402-aware: an agent caller signs a USDC payment over Base or Optimism mainnet and the tool returns the result with a signed payment-proof attached to the response envelope. Per-tool pricing is in the $0.05–$0.50 range — naming is a high-frequency, low-cost decision, and the pricing is set so a brainstorming agent can survey a hundred candidates for the cost of one human-hour.

The two surfaces share the same backend, the same scoring, and the same data. The website is the case where a human is the buyer. The MCP server is the case where an agent is the buyer. The product was built so the two cases are interchangeable — and that is the whole point.

Three design choices that put NameIntel on the right side of the autopilot economy

The agent is treated as a first-class buyer, not a footnote. The default posture across the SaaS world today is human-first: a web app with an account, an API as an afterthought. NameIntel inverts that — the MCP server is co-equal with the web surface, the x402 payment path is shipped on day one (not “we’ll add an API later”), and the tool descriptions are written for an LLM caller, not a developer reading docs. The bet is the same one Sequoia named: when the work moves from being done by humans to being done by software, the products that price the work — not the tool — are the ones that capture the value.

The pricing is per-query, not per-seat. A traditional SaaS naming tool would charge $19 per month for unlimited shortlists. NameIntel charges per call — cents per query, paid in USDC at the endpoint. The reason is mechanical: an agent doesn’t have a seat, doesn’t have a month, and doesn’t want a subscription it has to remember to cancel. It wants to pay for exactly the work it commissioned. Per-call pricing is the only price shape that fits the buyer; subscription pricing is a legacy of human-only commerce.

Honest unknown-states beat guessed answers. Several social platforms (Snapchat, Reddit in some configurations, X under certain rate-limit regimes) block unauthenticated availability probes. The right answer for those platforms is unknown, not a confidently-wrong available. The same discipline runs through the trademark, GEO, and SEO scoring: where the data is partial, the output says so, and the composite weighting reflects the missing surface rather than pretending it is covered. An agent caller will act on a wrong answer; the product’s job is to never give one.

The thesis behind the launch — Autopilot is the buyer change

NameIntel sits inside Theme #4 of the 30-day run — the Autopilot economy. The frame is the one Sequoia’s Services: The New Software essay named most clearly: when AI compresses the cost of doing the work, the buyer increasingly is the software doing the work, and the products that sell the work — not the tool — are the ones that scale. Day 9’s CogleGroup GP partnership was the operator form of that thesis: a services firm whose unit of delivery is an outcome, not a license. Day 18 is the product form: a service whose buyer is an agent, whose pricing is per-call, whose interface is an MCP tool, and whose settlement layer is a stablecoin micropayment over a public chain.

The long-form companion to this launch drops the same morning on Substack — #7: The Payment Is The Auth — x402, USDC, and the price shape an agent buyer actually wants. The essay is the payments-side argument for the buyer change; it pairs with #3 (the operator-side argument for what CogleGroup is doing on Sequoia’s autopilot map) and #6 (the engineering-side argument for the SDLC that built this run, which drops tomorrow to pair with Planwright).

Tomorrow — Day 19 — Planwright is the second leg of the same arc: the PM tool the agent-native SDLC needs, and the SOC 2 2026 audit-chain wedge. NameIntel and Planwright are not two unrelated products; they are two faces of the same buyer change. The two-week shape of the second half of the run is built on that change.

Pricing & business model

• Web — free preview, paid full scoring. A single dimension (domain availability) is free on the website. A full five-dimension composite is $0.99 per scored name on the web, paid via Stripe; an unlimited human-only tier is $19/month or $149/year for founders and naming consultants. The web tier is the side of the business that looks like a 2024 SaaS.
• MCP / x402 — per-call USDC mainnet, no subscription. check_domain $0.05/call · check_social $0.05/call · check_trademark $0.15/call · check_geo $0.20/call · score_name (all five) $0.50/call. Settled in USDC on Base or Optimism mainnet via x402. No account required; the payment is the auth. This is the side of the business that looks like the autopilot economy.
• Enterprise — naming firms and agencies. Custom volume pricing with a private MCP endpoint, white-label support, and CSV export of stored shortlists. Contact todd@silverbackcto.com. The reference target is the boutique-naming-firm shape — Lexicon, Catchword, Operative — where ten analysts run dozens of shortlists a week and NameIntel becomes the analysis substrate underneath the human creative.

The unit economics on the MCP side are deliberately built to clear margin even at low per-call prices: domain checks and social probes run against free or near-free upstreams; the trademark check uses the public USPTO API; the GEO score is the only paid input and it runs through Haiku at fractions of a cent per call. The composite scoring layer is essentially free at the marginal call. The MCP side becomes profitable on volume — which is exactly the call pattern an agent buyer produces.

The Velocity Process notes

What Claude Code handled: the Next.js 16 + React 19 + Tailwind v4 marketing and console site at nameintel.io — landing, console, pricing, the live result panels for each dimension, the example-name showcase, the llms.txt and robots.txt written in the product’s own voice; the TypeScript MCP server (mcp-nameintel) — the JSON-RPC 2.0 server, the five tool implementations with their input/output schemas, the tools/list discovery surface, the x402 gateway with the facilitator integration and the signed-payment-proof response envelope; the AWS CDK v2 stack on VelocityStack — API Gateway HTTP API v2, six Lambdas, two DynamoDB tables (scoring cache + payment ledger), Secrets Manager wiring for the Route53 / Bedrock / Stripe credentials, the monthly budget alarm; the per-dimension adapters (Route53 Domains + RDAP fallback for domain; USPTO TESS for trademark; the eleven-platform resolver for social; the deterministic-prompt SEO scorer; the Haiku-on-Bedrock GEO scorer); the GitHub Actions OIDC deploy pipeline with staging + prod environments.

What required human judgement: the two-surface decision — the simpler launch would have been the web app alone, with the MCP server promised for “later.” Shipping both surfaces on Day 1 is the entire argument of the launch; if NameIntel is just a website with an MCP server in the roadmap, the autopilot-economy positioning is a marketing line, not a product reality. The x402-mainnet-on-launch decision — the path of least resistance was a testnet-only x402 path at launch with a “mainnet this week” disclosure. The decision was to hold the launch through the weekend, debug the facilitator integration on Base and Optimism mainnet, and ship the launch with mainnet settlement actually live. The schedule slip is the cost of that decision; the integrity of the production-form argument the launch is making is the payoff. And the honest-unknown-state discipline — the instinct of a coding agent building this is to return a confident answer for every platform; the right call is to mark unknown where the upstream blocks probes, because an agent caller will act on a wrong answer and the product’s reputation is destroyed by exactly one falsifiable hallucination per ten thousand correct results.

What broke — and what the readiness pass caught: two integrity problems made it into the late drafts of the launch surface and were pulled before go-live. (1) The landing page carried a “Trusted by 50+ naming firms” social-proof line under the hero — invented, with no real firms to name. The Day 14 GEOPress and Day 15 Swole Labor Services readiness audits both caught the same class of failure (fabricated testimonials, an unearned approval badge), and the lesson is the same: placeholder layout is fine, placeholder claims are not. The line was replaced with a truthful “early access · agent-native pricing · USDC mainnet settlement” tri-fold. (2) The pricing page initially showed an “x402 verified” badge from a payment-protocol consortium that the project has not actually been certified by. Removed. The page now states the x402 implementation in plain terms — endpoint behavior, settlement chain, per-tool prices — without an unearned certification mark. The third issue the readiness pass surfaced was less an integrity break and more a scoping question: the MCP score_name tool’s GEO sub-score was returning slightly different rationales for the same input on back-to-back calls because the Haiku prompt was non-deterministic; the fix was to set temperature: 0, pin the system prompt, and cache the result by (name, prompt-hash) for a 30-day window. The same call now returns the same answer. The fourth — and the one that drove the schedule slip — was the x402 facilitator integration: the testnet path was clean by Friday afternoon, the Base-mainnet path lit up Sunday evening, the Optimism-mainnet path Monday morning. Shipping the launch with a testnet-only disclosure was the option we declined.

What I’d do differently: ship a public test harness on Day 1, not Day 3. The strongest proof for an MCP product is letting people see the tool calls work in their own runtime. A 60-second screen recording of score_name being invoked from Claude Desktop, embedded on the landing page, would convert better than any paragraph of copy. The work to produce that recording is small; the failure was treating it as marketing collateral instead of as part of the launch surface. The Day 19 Planwright launch should have its harness ready at the same moment the post goes out.

What’s next this week

• Day 18 (today): nameintel.io is live; the mcp-nameintel server is live; both x402 mainnet (Base + Optimism) and Stripe payment paths are wired and tested end-to-end; the WebPI submission for the MCP server registry is filed.
• Day 19 (Tue May 19): Planwright — the PM tool agents were waiting for; the SOC 2 2026 audit-chain wedge. Theme #4 (Autopilot economy) folds into the launch post as framing. Substack #6 (The SDLC When Intelligence Is Cheap) drops the same morning as the conceptual companion.
• This week: publish the example-shortlist case studies (with permission from the founders involved); ship the 60-second Claude Desktop demo recording; open the enterprise MCP-endpoint conversation with the two boutique naming firms in the warm list; monitor the x402 settlement chain for any reorg edge-cases on Base or Optimism mainnet; publish a per-chain settlement-throughput readout in Friday’s weekly recap.
• Monday (today): Substack #7 — The Payment Is The Auth — x402, USDC, and the price shape an agent buyer actually wants — drops the same morning as this launch. NameIntel is the production proof point the essay rests on; the essay is the long-form thesis the launch is the proof of.

Want to talk

If you run a naming firm or a brand agency — Lexicon, Catchword, Operative shape — and you have analysts shortlisting names by hand, NameIntel is the analysis substrate underneath the creative work. The enterprise path includes a private MCP endpoint, volume pricing, white-label support. Email todd@silverbackcto.com.

If you build agents that need to name things — a generative-product tool, a startup-naming assistant, an internal naming-bot for a product org — the MCP server is live at mcp.nameintel.io. Point your client at it, fund a wallet, and call the tools. The full schema is at nameintel.io/mcp. For volume or a custom arrangement, same email.

If you just need to score a name, the console is live at nameintel.io. The first dimension (domain availability) is free; the full composite is $0.99.

NameIntel is co-built with Jim Hasty and lives inside the autopilot-economy thesis the next two weeks of the Velocity Launch are organized around. Live today at nameintel.io. Tomorrow: Planwright — the PM tool agents were waiting for.