Get your GTM plan in 60 seconds
Paste your product URL. Bladesmith reads your site, writes an executable go-to-market plan, then runs the channels for you — and proves which actions moved the numbers.
Connect GitHub insteadFree tier · no credit card · nothing ships without your approval
- Diagnosis
- Visitors arriving from /pricing now complete onboarding 34% less often. The signup CTA moved below the fold in the new pricing layout.
- Proposal
- Restore the signup CTA above the fold on app/pricing/page.tsx and add the plan comparison anchor.
This is the whole product: a proposal, the evidence behind it, and one click.
The more you connect, the more it does
Bladesmith reads your site and returns positioning, ICP, prioritized channels and quick wins.
Unlocks · an instant GTM plan
GA4, PostHog or Search Console. Every suggestion is grounded in your real funnel instead of a template.
Unlocks · advice from your own numbers
Agents open pull requests and run the channels. Execution, not recommendations.
Unlocks · agents that ship
The shift
Producing marketing went free. The reader stopped being human.
Both halves of the world marketing software was built for ended in about eighteen months. A model writes the landing page, the launch post, the comparison table — and increasingly, a model is the one reading them.
69%
of searches end without a click
Up from 56% in a single year.
−61%
click-through where an AI answer appears
The rank you fought for is below the answer.
−38%
publisher referral traffic, year over year
The channel that carried software for 20 years.
90%
of B2B buying agent-intermediated by 2028
Gartner. Roughly $15T routed through agents.
Zero-click, click-through and referral figures: AI traffic decline, 2026. Agent-intermediated buying: Gartner, via Oro.
So a founder today has infinite output and zero judgment. You can generate five hundred things and not tell which one moved a number, because the loop that would tell you was never built.
Analytics answers what happened. Nothing answers which of my actions caused it, and what should I do next. That gap is the product. When making is free, the only thing worth paying for is knowing what to make.
The product
One loop, running every day, without being asked
Connect your product and your analytics. From that moment Bladesmith runs the same seven steps on repeat — and each turn of it leaves behind a fact about what works.
- 01
Signal
A number moves in your analytics — or stops moving.
- 02
Diagnosis
It reads the funnel, the site and the code to work out why.
- 03
Proposal
A specific action, with the evidence, the expected effect and the cost.
- 04
Approval
You approve in one click. Nothing happens before that.
- 05
Execution
A pull request, a send, a write to a connected tool — whichever route is available.
- 06
Outcome
The metric is measured after the fact and attached to the action that caused it.
- 07
Learning
The result updates what we recommend next — for you, and for products like yours.
The founder's surface is an inbox: ten minutes a day, approving work that has already been reasoned about.
See the loop in detailExecution
It executes in two places
Some of the highest-leverage marketing for a software company isn't marketing at all — it's code. The rest happens out on the channels. Same loop, two surfaces.
Inside your product
As pull requests
The landing page, the pricing copy, the onboarding flow, the structured data that makes you readable by assistants, the instrumentation that measures any of it. Bladesmith writes them as diffs you review and merge — auditable, revertible, in the tooling you already use.
A pull request is the best consent mechanism ever designed.
export default function Pricing() {return (- <section className="pt-24">- <PlanGrid plans={plans} />+ <section className="pt-12">+ <SignupCta id="start" variant="primary" />+ <PlanGrid plans={plans} compareAnchor /></section>
Out across the channels
As a fleet of specialist agents
Each agent owns a channel end to end: its rules, its register, its timing. It finds the moment, drafts what belongs there, and executes on approval — or hands you the finished asset when that channel has no safe automated route.
- SEOexecutes
- Cold emailexecutes
- Contentdrafts
- Redditdrafts
- Product Huntdrafts
- Review sitesdrafts
- Review requestsexecutes
- Marketplace adsdrafts
Proof
An action you can't measure hasn't been executed. It's been performed.
Every action Bladesmith takes gets an outcome attached to it: the metric, the movement, the window. This is the hard half that the rest of the category skips — and the only reason the advice gets better.
Connected in a click. If you have none of them, the first pull request is the instrumentation.
Cumulative measured outcomes
Illustrative- 1Instrumented the funnelbaseline set
- 2Shipped the comparison page+14% organic
- 3Rewrote the pricing CTA+8.1% activation
The Outcome Graph
Your result improves the next founder's first guess
Every closed loop leaves a fact behind: for a product of this shape, at this stage, this action moved this metric by this much. One is an anecdote. Ten thousand, connected and anonymised, is the only empirical prior on go-to-market that has ever existed.
Products, linked by learned similarity
IllustrativeIt is cross-project
Your measured outcome improves the first recommendation the next founder gets, and theirs improves yours — a real network effect, not a data-volume claim.
A model cannot have it
A frontier model will out-write us forever, and we use it for exactly that. It will never know which of two onboarding emails converted for the four hundred products most like yours. That fact was never published.
It is safe to be part of
Anonymisation thresholds and approval gates sit between your project and the graph. What travels is the shape of a result, never anything identifiable.
Both sides of the exchange
Your next buyer may never load your page
Assistants already mediate discovery, and the commerce rails underneath them are being built now. When both ends are machines, marketing stops being persuasion and becomes an interface problem: can a machine find you, read you, compare you, and buy from you?
- Structured product facts, llms.txt, JSON-LD and an MCP endpoint, so assistants answer about you correctly instead of guessing.
- Share of model: how often you are named, for which questions, against whom.
- The agent-to-agent channel itself — and the commerce rails forming under it.
Share of model
How often an assistant names you in its answer
Where we stand
They measure. We close the loop.
The market offers four positions. Three of them stop before the part that compounds.
| Position | Typical product | Where it stops |
|---|---|---|
| Generate | AI writing assistants | Sells the one input whose price is collapsing. A product whose value is a prompt has a shelf life. |
| Publish | Schedulers, AI SDR tools | Automates the sending, not the deciding. Volume without judgment is the problem, not the fix. |
| Measure | AI-visibility dashboards | Right about the problem. It tells you that you are invisible — it cannot make you visible, or prove a change worked. |
| Close the loop | Bladesmith | Measures, acts, proves the action moved the metric, and learns across projects which action to try next. |
A dashboard is a product; a loop is a flywheel. Closing it requires execution rights — a repo, a key, a send — and nobody hands those to a stranger. That difficulty is the moat, and its entry fee is trust, which is why every single thing we do is proposed, reviewed and revertible.
Pricing
Free intelligence. Paid hands.
We never charge for the part that is going to zero. The plan, the analysis, the diagnosis and the assets are free — permanently and generously. You pay for executed actions, and only when they succeed.
Free forever
The GTM plan, the site and funnel analysis, the diagnosis, the Growth Kit and every ready-to-use asset. Rung zero — “here is the finished thing, do it yourself” — is always available.
Metered
Executed actions: pull requests opened, emails sent, writes to your connected tools — plus advanced memory and the cross-project learning layer. Charged per action, on success.
Questions
The ones we actually get asked
Do I have to give you access to my repository?
Only if you want execution. The plan, the analysis and the assets work from a URL. When you do connect a repo, every change arrives as a pull request you review, merge and can revert — we do not ask for trust, we ask for a review.
What if I have no analytics installed?
Then the first thing Bladesmith proposes is the instrumentation: a pull request that wires up your funnel — acquisition, activation, conversion — so everything after it can be measured. An action whose outcome cannot be observed has not been executed, it has been performed.
Will an agent post or send anything on its own?
No. Every action stops at your approval, on every channel. Where a channel has no safe automated route, the agent hands you the finished asset instead of pretending — it never blocks and never bluffs.
What happens when the models get better?
We get better. Drafting quality rises for free, cost per action falls, and the ceiling on what an agent can safely execute rises — which turns advice into action. Our asset is evidence about what worked, and models consume evidence rather than producing it.
What crosses the boundary between projects?
The shape of a result — this kind of action moved this metric for this kind of product — behind anonymisation thresholds. Nothing identifiable: not your code, not your customers, not your copy.
Who is this built for, right now?
Founders who ship from a repo — Cursor, Claude Code, Lovable — and have no marketing team. It is the one segment that already lets an agent write to production, which is why the loop can close there first.
Your product is built. Now it has to be found.
Paste your URL and read your go-to-market plan in about a minute. Connect the repo when you want the agents to start shipping.
Connect GitHub insteadFree tier · no credit card · nothing is sent, posted or shipped without your approval