Shane Logsdon
@shane.logsdon.io
Founder @ LeadSurface · Surfacing in-market buyers from the forums where they’re already talking · Developer Advocacy @ Global Payments · buying signals, intent data
The most-shared AI implementation playbook gets the order right: data, then workflows, then intelligence. It waves at the hard parts in two sentences, test sets and logging, and that's where implementations are won and lost. I annotated both. shane.logsdon.io/articles/str...
September 26, 2026 at 8:30 PM
The most-shared AI implementation playbook gets the order right: data, then workflows, then intelligence. It waves at the hard parts in two sentences, test sets and logging, and that's where implementations are won and lost. I annotated both. shane.logsdon.io/articles/str...
261 skills were sitting in my Pi configuration. 187,801 characters of catalog, on every request. I moved it out and made it searchable, so skills stay on disk until a task calls for one.
September 21, 2026 at 1:05 AM
261 skills were sitting in my Pi configuration. 187,801 characters of catalog, on every request. I moved it out and made it searchable, so skills stay on disk until a task calls for one.
A metric that goes up when you're less necessary is a broken metric. Docs views and stars now measure how well AI replaced your content, not your program. The real question: do you show up when someone asks ChatGPT what to use? shane.logsdon.io/articles/str...
September 20, 2026 at 4:30 PM
A metric that goes up when you're less necessary is a broken metric. Docs views and stars now measure how well AI replaced your content, not your program. The real question: do you show up when someone asks ChatGPT what to use? shane.logsdon.io/articles/str...
Answer engines can't see behind your signup form. Gate the whitepaper, and the AI your buyer just asked never gets to cite it. Value creation used to be the idealistic long game. Now it's the only way to show up in the answer.
shane.logsdon.io/articles/str...
shane.logsdon.io/articles/str...
September 19, 2026 at 8:30 PM
Answer engines can't see behind your signup form. Gate the whitepaper, and the AI your buyer just asked never gets to cite it. Value creation used to be the idealistic long game. Now it's the only way to show up in the answer.
shane.logsdon.io/articles/str...
shane.logsdon.io/articles/str...
I audited Stripe, Clerk, and Neon's docs for AI readability. All three shipped an llms.txt. None of them got past layer two of a six-layer stack. Nobody owns this work yet, which means DevRel can.
shane.logsdon.io/articles/tec...
shane.logsdon.io/articles/tec...
September 18, 2026 at 4:30 PM
I audited Stripe, Clerk, and Neon's docs for AI readability. All three shipped an llms.txt. None of them got past layer two of a six-layer stack. Nobody owns this work yet, which means DevRel can.
shane.logsdon.io/articles/tec...
shane.logsdon.io/articles/tec...
"Should this AI workload run locally or in the cloud?" For a small business the honest answer is usually "it depends."
I'm starting a public research project to get to something better. 🧵
I'm starting a public research project to get to something better. 🧵
September 16, 2026 at 11:56 PM
"Should this AI workload run locally or in the cloud?" For a small business the honest answer is usually "it depends."
I'm starting a public research project to get to something better. 🧵
I'm starting a public research project to get to something better. 🧵
A workflow that never throws an error can still be wrong for weeks. Broken is loud. Wrong runs clean and drifts quietly until someone finally asks why the output got worse. shane.logsdon.io/articles/tec...
September 14, 2026 at 4:30 PM
A workflow that never throws an error can still be wrong for weeks. Broken is loud. Wrong runs clean and drifts quietly until someone finally asks why the output got worse. shane.logsdon.io/articles/tec...
AI fakes any tutorial. It can't fake a dead end. Witnessed practice, watching someone competent work through real uncertainty, is the DevRel format AI can't replicate. Corporate DevRel barely makes any. shane.logsdon.io/articles/str...
September 13, 2026 at 4:30 PM
AI fakes any tutorial. It can't fake a dead end. Witnessed practice, watching someone competent work through real uncertainty, is the DevRel format AI can't replicate. Corporate DevRel barely makes any. shane.logsdon.io/articles/str...
Stack Overflow's down to 2009 question volume, and it's tempting to read that as community dying. It isn't. Q&A was never the community. The people who kept showing up after they stopped being stuck were, and AI made that impossible to miss. shane.logsdon.io/articles/ind...
September 12, 2026 at 8:30 PM
Stack Overflow's down to 2009 question volume, and it's tempting to read that as community dying. It isn't. Q&A was never the community. The people who kept showing up after they stopped being stuck were, and AI made that impossible to miss. shane.logsdon.io/articles/ind...
I ran my own citation test on myself. 0% when I asked AI what to use for my category, and 100% when I asked who I am. Every assistant knows my name, but none of them recommend me for what I do. shane.logsdon.io/articles/tec...
September 11, 2026 at 4:30 PM
I ran my own citation test on myself. 0% when I asked AI what to use for my category, and 100% when I asked who I am. Every assistant knows my name, but none of them recommend me for what I do. shane.logsdon.io/articles/tec...
Nobody builds preference because your quickstart explained pagination clearly. AI does that instantly. What's left for DevRel is judgment: the architecture calls and tradeoffs a model can't make for you. shane.logsdon.io/articles/lea...
August 12, 2026 at 4:30 PM
Nobody builds preference because your quickstart explained pagination clearly. AI does that instantly. What's left for DevRel is judgment: the architecture calls and tradeoffs a model can't make for you. shane.logsdon.io/articles/lea...
AI coding agents read your docs before any developer does, and no analytics metric captures it. Bounce rate, session depth, and page views stop meaning what they used to when the reader never loads your site. shane.logsdon.io/articles/str...
August 1, 2026 at 4:30 PM
AI coding agents read your docs before any developer does, and no analytics metric captures it. Bounce rate, session depth, and page views stop meaning what they used to when the reader never loads your site. shane.logsdon.io/articles/str...
Finally wrote up the AI system I use to build my side projects in the cracks of a full life. Turned it into three free kits for building, marketing, and staying honest about follow-through. Public, MIT, free. You just need a paid Claude plan.
July 14, 2026 at 8:57 PM
Finally wrote up the AI system I use to build my side projects in the cracks of a full life. Turned it into three free kits for building, marketing, and staying honest about follow-through. Public, MIT, free. You just need a paid Claude plan.
Where do your ideas go when they show up at a red light? The bug fix and the onboarding wording never arrive at the computer. Forty minutes of building time a day means the real work is catching the thought before it slips.
shane.logsdon.io/articles/str...
shane.logsdon.io/articles/str...
Building on the Margins
A founder with a day job and family builds LeadSurface via a Loop & Gate workflow, keeping product strategy human, handing disciplined execution to a pipeline.
shane.logsdon.io
July 13, 2026 at 5:51 PM
Where do your ideas go when they show up at a red light? The bug fix and the onboarding wording never arrive at the computer. Forty minutes of building time a day means the real work is catching the thought before it slips.
shane.logsdon.io/articles/str...
shane.logsdon.io/articles/str...
A workflow that runs without errors can still be failing your core objectives through silent drift. You must measure the quality of intent, not just execution speed. shane.logsdon.io/articles/tec...
June 22, 2026 at 9:02 AM
A workflow that runs without errors can still be failing your core objectives through silent drift. You must measure the quality of intent, not just execution speed. shane.logsdon.io/articles/tec...
The wrong workflow is harder to find than the broken one. It runs cleanly and produces output that looks right, but the error accumulates over time. shane.logsdon.io/articles/tec...
June 19, 2026 at 2:27 AM
The wrong workflow is harder to find than the broken one. It runs cleanly and produces output that looks right, but the error accumulates over time. shane.logsdon.io/articles/tec...
The dispatch layer rewrites short input into a detailed prompt and routes it to the right agent. Each agent is just a system prompt plus a model alias, so adding one is easy. DIY, guided, or Docker setup.
github.com/slogsdon/her...
github.com/slogsdon/her...
June 6, 2026 at 2:49 PM
The dispatch layer rewrites short input into a detailed prompt and routes it to the right agent. Each agent is just a system prompt plus a model alias, so adding one is easy. DIY, guided, or Docker setup.
github.com/slogsdon/her...
github.com/slogsdon/her...
Claude Dispatch gave me the mental model, then I wanted a local version on my own hardware. So: hermes-dispatch. Type a request on your phone, it routes to one of 24+ local agents via Ollama + LiteLLM. Over Tailscale, nothing leaves your network. MIT.
github.com/slogsdon/her...
github.com/slogsdon/her...
June 6, 2026 at 11:49 AM
Claude Dispatch gave me the mental model, then I wanted a local version on my own hardware. So: hermes-dispatch. Type a request on your phone, it routes to one of 24+ local agents via Ollama + LiteLLM. Over Tailscale, nothing leaves your network. MIT.
github.com/slogsdon/her...
github.com/slogsdon/her...
Ask ChatGPT "best [your service] in [your city]."
If you're not in the answer, it's not your Google ranking. Five structural signals separate you from being the citation, and most local sites have none of them.
shane.logsdon.io/strategic-in...
If you're not in the answer, it's not your Google ranking. Five structural signals separate you from being the citation, and most local sites have none of them.
shane.logsdon.io/strategic-in...
May 26, 2026 at 1:02 PM
Ask ChatGPT "best [your service] in [your city]."
If you're not in the answer, it's not your Google ranking. Five structural signals separate you from being the citation, and most local sites have none of them.
shane.logsdon.io/strategic-in...
If you're not in the answer, it's not your Google ranking. Five structural signals separate you from being the citation, and most local sites have none of them.
shane.logsdon.io/strategic-in...
I thought the AI was failing. It was doing exactly what I'd set it up to do.
Every output gap traced back to a decision I hadn't made. What counts as a failed fetch versus an incomplete one? Which fields are required? Who owns retry behavior? I hadn't decided, so the agent decided for me.
Every output gap traced back to a decision I hadn't made. What counts as a failed fetch versus an incomplete one? Which fields are required? Who owns retry behavior? I hadn't decided, so the agent decided for me.
May 13, 2026 at 12:35 AM
I thought the AI was failing. It was doing exactly what I'd set it up to do.
Every output gap traced back to a decision I hadn't made. What counts as a failed fetch versus an incomplete one? Which fields are required? Who owns retry behavior? I hadn't decided, so the agent decided for me.
Every output gap traced back to a decision I hadn't made. What counts as a failed fetch versus an incomplete one? Which fields are required? Who owns retry behavior? I hadn't decided, so the agent decided for me.
68 of 124 skills in my Claude Code setup had never fired.
I audited every one. Mined six weeks of session transcripts. Scored each on trigger clarity, instruction quality, output spec, and anti-pattern density. Then ranked them by composite score against actual usage.
I audited every one. Mined six weeks of session transcripts. Scored each on trigger clarity, instruction quality, output spec, and anti-pattern density. Then ranked them by composite score against actual usage.
May 9, 2026 at 12:23 AM
68 of 124 skills in my Claude Code setup had never fired.
I audited every one. Mined six weeks of session transcripts. Scored each on trigger clarity, instruction quality, output spec, and anti-pattern density. Then ranked them by composite score against actual usage.
I audited every one. Mined six weeks of session transcripts. Scored each on trigger clarity, instruction quality, output spec, and anti-pattern density. Then ranked them by composite score against actual usage.
You're still designing for yourself.
AI coding tools don't fail on prompts. They fail on ambiguous artifacts. When the spec is vague, the agent fills the gaps with what it was trained to assume, not what your system actually needs.
AI coding tools don't fail on prompts. They fail on ambiguous artifacts. When the spec is vague, the agent fills the gaps with what it was trained to assume, not what your system actually needs.
May 3, 2026 at 5:23 AM
You're still designing for yourself.
AI coding tools don't fail on prompts. They fail on ambiguous artifacts. When the spec is vague, the agent fills the gaps with what it was trained to assume, not what your system actually needs.
AI coding tools don't fail on prompts. They fail on ambiguous artifacts. When the spec is vague, the agent fills the gaps with what it was trained to assume, not what your system actually needs.