Blog
What we are learning building AI-native growth systems for consumer and software brands. Pieces from inside the work, plus analysis of the AI-native services landscape we track every day.
An AI conversion tool's edge is speed — more tests, faster. But more tests mechanically manufacture more false winners, so a pile of reported lifts can grow while your actual conversion rate stays flat. The one check that catches it, and two moves the tool won't hand you.
Read moreMeta Advantage+ and Google Performance Max run the buying now — and the same machine ships to every advertiser. How to feed it what the platforms themselves ask for, and build the outer loop: the record of why winners won that compounds for you alone.
Read moreEvery CRO program stalls on the same handful of problems. AI solves some, only speeds up others, and can’t touch two — the traffic floor and where your learnings live. Sort yours before you buy a tool.
Read more
Two brands spend the same for years; one’s marketing compounds, the other resets every campaign. The difference is entity propagation — define each thing your brand is known for once, and carry that exact definition across every surface.
Read moreA new class of AI conversion tool doesn't just run your A/B test — it writes the variants too. Here's what quietly changes when one tool both writes the copy and grades it, and the three things to lock before you turn it on.
Read moreType your company into ChatGPT and a model hands your buyer a description you never wrote. An llms.txt won’t fix it — your ordinary pages are the training data. State what you are in checkable terms.
Read moreWe run an AI-native demand-creation engine on TheNextGuide. Here's the real operating week — what it makes, where it lands, and the one thing the team approves that turns a content calendar into a system that compounds.
Read moreThe agency lead magnet is escalating from gated PDF to copy-paste prompt to run-for-you audit widget — each rung pricing the diagnostic at zero. The test for what survives: if your diagnostic fits in a prompt, you're selling whatever you do after it.
Read moreThe fourth part of a brand-install is the memory layer — what a firm learns about your brand, kept with the reason. The methods transfer between brands; the memory never does, which is why it’s the one part that’s truly yours when the engagement ends.
Read moreAn AI conversion tool that moves traffic to the leader mid-test can't hand you a clean lift. Here's why, the read the platforms document but hide, and three questions to ask before you bank an "uplift."
Read moreThe Learn step of an AI content engine: when to measure (a fast beat for anomalies, a slow beat for decisions, cadenced to your volume), what to measure (signals that survive small samples), and how the loop reallocates toward what won. A number you never act on is a scoreboard, not a feedback loop.
Read moreSatya Nadella says the model isn't the moat — the learning loop you own is. The one sovereignty question to ask any AI vendor before you sign: swap the model, do you keep what it learned?
Read moreAdding 'an AI review step' isn't safety — a checker that fails for the same reasons as the writer shares its blind spots. How defense in depth (independent layers) and risk-tiering let an AI content engine ship fast without shipping mistakes.
Read moreTwo brands can post the same revenue curve — one built a flywheel, one rented a treadmill. The one number that tells them apart is the fully-loaded cost to produce a unit of growth, watched as a trend — falling means compounding, flat under rising spend means rented.
Read moreAn AI-native install can only act on the customer it can see — and on day one that customer is scattered across five tools that disagree. Why the real first phase of the work is data archaeology, not agents.
Read moreRalph Lauren gave its AI shopping assistant a press release and its forecasting AI one sentence. The press-release test shows a DTC founder which kind of AI investment their brand is actually making.
Read moreWe read ten AI conversion-tool homepages. The pitch converged on one promise: it runs itself. What it stopped answering is where the hypothesis comes from — and these three steps your team keeps in-house decide whether any tool pays off.
Read moreThe autonomous-agent wave sells the removal of the human. But a brand letting AI act on its accounts is buying outcomes it can stand behind — and that takes an accountability layer on day one: scoped access, an audit log, approval gates, and a kill switch the brand holds.
Read moreA founder ran her quarterly retention review in April and learned about January's churn too late. Retention stopped being a calendar and became a loop — three structural shifts AI exposed, with a way to tell which side of each your stack is on.
Read moreThree unrelated sources settled in eleven days what gets a brand cited by AI search. We sampled nine GEO/AEO vendors to see what they sell instead — receipts included.
Read moreYour dashboard sees the 10% that clicks. The five-signal stack measures the rest — self-reported answers, brand search, the founder’s calendar, community inbound, and a weekly pass that reads them together.
Read moreAI agents now read your store before your customers do — and the traffic they hand back converts 42% better. What changes in a CRO sprint when the conversion surface has two audiences: humans clicking and agents fetching.
Read moreMost agencies keep their method in a senior person's head or a slide deck — neither compounds. The third option is a library of small instructions files that run: capture a task once, generalize it, reuse it across brands. Garry Tan runs his own work this way; here's the same shape moved into a services firm, and why the growing library is the asset.
Read moreTwo five-layer frames apply to content. Blomfield's loop is how the work gets smarter; Birkett's surface — Strategy, Enablement, Execution, Feedback, Repurposing — is what a small team has to staff. Here's the operating model that compounds, and the layer most teams under-staff.
Read moreDavid Mannheim's reframe from Speero Circus 2026 — designing experiments around intent, not pages — operationalized as a five-step research loop with AI at the upstream read layer and the team picking the fights.
Read moreAI-native services delivery has had a stable engineering vocabulary in public for more than a year. Most services firms still describe AI as a capability bolted onto existing services — and the bolt-on framing without that vocabulary underneath is one of the strongest audit signals a buyer has.
Read moreWe're moving TheNextGuide's content engine from automated to AI-native. The line isn't the model — it's the feedback loop. Here's the seven-step install at a tourism marketplace we operate, the structural difference named, and four questions to audit any AI marketing tool in your stack.
Read moreMost B2B SaaS teams fund the half of the funnel their attribution dashboard can see and underfund the half it can't. Three signals you're on the wrong side of the flip, the AI-native rebuild of Walker's demand-creation framework, and a 30-day on-ramp. Series A (AI-Native Demand Creation), post 1 of 4.
Read moreWhat does an AI-native services firm actually hand a client? Not a team — a brand-install: a folder of plain text the AI reads on every task. Five components (working hypothesis, locked rules, instructions index, memory index, kill switch), one weekly self-improvement loop, and the moat that lives in the catalog. Series B (Codified Engagement), post 1 of 4.
Read moreThe companion install guide to the self-improving content engine: Blomfield's five layers as plain-text files in a folder, the week-by-week install order across the first month, the routing that loads the right slice per task, and the four failure modes that break it.
Read moreMost companies bolt AI onto an old workflow and call it productivity. The reframe: redesign the workflow as a recursive five-layer loop that gets smarter every Friday — sensor, policy, tool, quality gate, learning mechanism. Here's the loop we run this blog through.
Read moreService as Software has reached legal, insurance, healthcare, tax, IT, and security — but not marketing. The structural reason: coding's repository made its work product legible by convention; marketing's never did. Here's the gap blocking it — and the per-client artifact that closes it.
Read moreAt least one AI tool in your stack won't exist in its current form 18 months from now. The company is doing one of three things — rebuilding from scratch with the agent doing the work directly, getting quietly absorbed into a larger platform, or falling behind by bolting AI features onto older software. The careers page tells you which, months before any press release. Five-minute audit at the end.
Read moreA founder I work with sat down with last quarter's attribution data and spent two days trying to explain why three of her five biggest deals had "direct" as the source. The dashboard wasn't broken — the buyer's journey moved. Three structural shifts in acquisition that AI didn't cause but exposed, with the test for whether your stack is on the right side of each.
Read moreAt Speero Circus 2026, three independent talks landed on the same shape: AI generates knowledge fast, but unreviewed AI output becomes organizational truth. The discipline AI-native marketing teams are converging on is a pull-request culture for knowledge — labeled reviewer, pass/fail signal, back-edit before merge.
Read moreTwo DTC brands run CRO for three months. One turns on an AI vendor across hero, PDP, cart, and checkout; the dashboard shows compounding lifts. The other runs eight customer interviews, finds one bad piece of copy, fixes it, and moves the same number in a single week. Both 'did CRO.' Only one ran the discipline auto-tuning vendors charge to skip.
Read moreA loyal customer called support twice last week and waited 38 minutes for a callback. Your email tool doesn't know — so the AI agent sitting on top of it just sent her the win-back discount. The agent didn't get it wrong; the data it could see was wrong. Before you point an AI at your marketing stack, run the audit that's upstream of the install.
Read moreRize published the Micro Agency Playbook — five tenets that read like a checklist someone wrote after watching the same shape emerge across Every, Crosby, Pace, Butler/Till, and the Forward Deployed Engineer pattern. None of these companies set out to be a Micro Agency. They just kept making the same architectural choices. Leanboat is a Micro Agency for consumer and software brands.
Read moreGarry Tan calls it the difference between a filing cabinet and a nervous system. The same architecture works at brand scale — and most brands can't install it alone. The real test of an AI-native engagement: what's still running in the brand's repo six months after the agency stops billing.
Read moreTwo operators on the same team, same model, same brief, ship wildly different output. The gap is not prompt-craft. It is delegation discipline — the ability to write down what you are thinking cleanly enough that another person, or an agent, can act on it. Here is how to spot a bad delegator before the AI install.
Read moreThree AI-native services firms in three different verticals are publishing the same artifact in 2026 — a named operating method, in writing, on the public site. The naming is not a brand-voice flourish. It shortens the sales cycle, pre-filters operator hires, justifies productized pricing, and makes the firm defensible against agencies that cannot be lined up on a comparison page.
Read moreMost 'AI for marketing' rollouts in 2026 feel like working for a chat window — the operator types, edits, ships, repeats, and the work never compounds. The fix is not a better model. It is writing the function down well enough that an agent can run it. Two installs from inside our own work, plus the same shape applied to B2B SaaS.
Read moreAI on top of paid acquisition compounds whatever the destination already does. If the funnel converts, the lift is real. If it does not, AI just makes the leak faster. The CRO install we run before — or alongside — the rest.
Read moreDeep-dive on box 1 of the AI-Native Foothold framing. The senior person's most repetitive five hours a week, automated. 30-day install, real math, predictable failure modes named upfront.
Read moreDeep-dive on box 2 of the AI-Native Foothold framing. AI handles the first response on customer queries, sales leads, and internal questions. Humans handle everything else. The line is sharp on purpose.
Read moreDeep-dive on box 3 of the AI-Native Foothold framing. Turn what lives in your team's heads into a queryable system before someone leaves. The slowest install to feel and the one that compounds the most.
Read moreDeep-dive on box 4 of the AI-Native Foothold framing. AI between the junior output and the senior reviewer. The bottom of the team's output rises to meet the top; the senior person catches what only they can.
Read moreMost owners want AI in their company. They cannot articulate where it fits. Generic training does not fix that. There are four jobs that pay back in 30 days. Pick one. Install it. Then pick the next.
Read moreFive questions to run on the next agency that pitches you AI. AI-painted agencies fail multiple questions in a row. AI-native agencies pass all five with specifics. About 30 minutes of the prospect's time.
Read moreThe published 5x marketer numbers are real. What gets skipped is the three-layer stack underneath — research, methodology, process — that has to all be present for the math to work.
Read moreTwo thousand years from the Roman contubernium to your current org chart, the constraint hasn't moved. Block just published the first public-company manifesto for what comes after — and the implication for everyone smaller than Block matters more than the headline did.
Read moreMost retention programs run as open loops — work goes out, metrics come back, the system never improves between attention. Closing the loop is now possible at small-brand scale. Here is what changes when it happens.
Read moreIn one quarter, six AI-native services companies hit valuations totalling north of $20 billion. The architectural pattern underneath all six is the same — and marketing is the lagging vertical.
Read moreSix independent publications now describe the same shape for an AI-native marketing team — three roles plus AI agents underneath. The framework is right at one scale and wrong at another. Here is both.
Read moreEvery employs 15 people, ships five products, runs a daily newsletter and a $1M consulting business, and writes 100% of their code with AI. The architecture is documented. Here is what 15 people now ship.
Read moreThe category exists now. Six months ago AI-native agency was a thesis; today there are published exemplars and a documented playbook. We are an AI-native agency for consumer and software brands. Here is what that actually means.
Read moreThe rule for every piece of content we ship: use AI to move faster around things only you can say. Don't use it to produce things anyone could say. Lily Ray has watched 20–30 sites lose traction in three months ignoring this rule.
Read moreMost agencies pitching AI in paid are pitching the wrong AI. Ad copy variations are the small play. Agentic media buying that closes the loop between bid, creative, audience, and outcome is the architectural shift.
Read more