Blowing It Up Daily, Building It Back Better

Let’s set the scene: I’m on stream, shoulder-deep in popcorn, and I realize halfway through a sentence that I didn’t pay for the subscription I need. Classic.

Welcome to The Sarah Factor, where we build, break, and rebuild in public—because that’s the only way the work gets better. If Morning Scrum is about shipping, The Sarah Factor is about the philosophy behind shipping: the daily ritual of blowing things up and putting them back together with slightly more soul.

Today’s episode is about Sarah—aka Signal—the AI SDR experiment I’m building at strategist.com/signal. It’s the usual Govind trifecta: picking tools, ditching tools, and figuring out where the human stops and the machine starts. If you came for a glossy pitch, wrong room. This is about the texture—the real “we’re in it” version—of building outbound that actually works.

The AI SDR with a Heartbeat

The project is straightforward on paper: build an AI-assisted SDR workflow that doesn’t feel like spam, doesn’t collapse into a compliance issue, and keeps the founder in the loop. The twist: do it with tools that don’t fight you and a philosophy that expects your stack to change every week.

Sarah/Signal is exactly that. It’s not just “AI sends messages.” It’s a system that pulls documents, PDFs, LinkedIn posts, personas, and campaign parameters into one living workspace, generates ultra-specific outbound, and then slows down at the right moments so a person can sanity-check before anything hits the wire.

If you’re imagining a wall of dashboards and a dead CRM, you’re still playing 2018. The game now is fluidity paired with auditability. Forget “set and forget.” Think “set, run, observe, edit, rerun.” Every pass makes the system sharper, lighter, more obvious. You’re not writing a script; you’re tuning an instrument.

Blow It Up, Then Make It Art

Here’s the core methodology: every day, blow it up and put it back together. If the pipeline can’t withstand daily reassembly, it’s too brittle to scale. That’s not cute rhetoric—this is how you avoid operational lock-in long before you fear vendor lock-in. I tried Zoho for the CRM layer, LinkedHelper for the automation, and within two days I threw them out. Not because they’re “bad”—they’re fine—just not philosopher-proof. As soon as an experiment required a left turn, I hit friction. If the system resists your curiosity, you walk. Curiosity pays the bills in this economy.

Tools aren’t just “features.” They’re cultural commitments. The wrong tool shapes your habits, nudges you away from nuance, and suddenly your outbound looks like everyone else’s. Mass-market automation loves sameness. The Sarah Factor hates it.

Attio: Finally, a CRM That Thinks in Data

Then I found Attio. Immediate sigh of relief. It’s what a CRM looks like when someone designs from the data outward rather than from a 2008 sales manager’s wish list. There’s a simplicity that invites tinkering. Custom fields aren’t a dungeon; they’re first-class citizens. The LinkedIn sidebar is actually usable: nudge open a profile, see what matters, log context without feeling like you’re feeding a compliance machine.

A CRM should be a graph you can think through, not a black hole with fields. The elegance here is that Attio doesn’t try to be the workflow. It’s the source of truth. That’s the right division of labor. Keep your workflows in the places where iteration is cheap; keep your truth where the schema is coherent.

SyncIn: The $30 That Saves Your Brain

If you’ve ever tried to sync LinkedIn messages into a CRM, you know the horror: one big blob of conversations dumped into a single note, stripped of context, fodder for no one. SyncIn fixes this. Thirty bucks a month, plug in, and your messages actually land where they belong, correctly threaded, in Attio, without you writing a crawler that breaks every third Thursday when LinkedIn sneezes. It sounds small, but this is the difference between a living system and a spiritual tax.

This is the recurring theme: the tools don’t need to be epic; they need to disappear. When you find a tool that solves a painful, boring problem completely, pay for it and never think about it again. Bandwidth is more expensive than SaaS.

Replit: Vibe Coding Meets the Credit Meter

Let’s talk about the AI Credit Economy because this part is wild. I love Replit for prototyping. Vibe coding is real: open a tab, summon a notebook, ask the assistant to spin up a workflow, and you’ve got an end-to-end automation pretty fast. It’s like pair programming with a ghost who knows the docs and isn’t jittery about semicolons.

But then I realized I ran out of credits. Worse: hovering over a suggestion, just mousing around a code change, was costing me about a dollar a minute. Not a joke. Just thinking was metered. That’s when it clicked: the running joke in AI is going to be about credits. Everything is consumption. A cloud provider’s dream state. Your attention, their unit price.

Is that bad? Not necessarily. It’s honest. Pay for the work you invoke. The danger is forgetting you’re paying for thought. The defense is clarity: define what work your agent should do, keep sessions tight, and don’t let tools swag around in the background because “it feels productive.” In the AI world, “vibes” can invoice you.

Q’s: The Holy Grail Workspace, Minus the Hype

So I rebuilt in Q’s. Q’s is the holy grail of internal AI workspaces for one very specific reason: clean context without ceremony. You create a folder, drop PDFs, documents, LinkedIn posts, cheat sheets, process docs—everything the system needs to think—and you build right there. You don’t need a public web deployment to be useful. You want a controlled, deterministic space where your system has context and your operators have clarity. That’s Q’s.

This is also where model opinionation goes to die. I am allergic to LLM lock-in. Pick the model per job. You want visuals? GPT’s got the aesthetics. You want conversational elasticity? Gemini’s great at communication. Claude’s image output? Not my favorite. Text nuance between Gemini 3 Flash and 3.1 Pro? Noticeable. Don’t marry your model; date them all. Loyalty to a provider is not a virtue if your product performance depends on switching lanes midstream.

Q’s makes that choice easier because the workspace isn’t married to one LLM. You can swap, test, and control the narrative with a folder-centric approach. It means the system is about the story—the documents, the examples, the data—and the model is a choice, not a fate.

ConnectSafely API: Automation With a Seatbelt

Mechanically, for LinkedIn automation, I’m using the ConnectSafely API. This keeps things sane. It’s enough to handle the templated work—the searches, the lightly structured outreach, the status changes—without unloading a bot army on the world. And because it’s API-based, I can layer my human-in-the-loop logic on top without living in browser automation hell. It’s the right abstraction: automate the boring parts, preserve human judgment where it matters.

A good rule of thumb: if a step can be robustly defined and audited, automate it. If a step is reputational—tone, fit, timing—pause and let a human stare at it for five seconds. That’s the whole outbound game right now. Five seconds at the right moment is the difference between a blocklist and a lunch.

Human-in-the-Loop: “Gary, Let’s Have Lunch…”

I don’t believe in fully automating outbound. Not because it can’t be done—the tech’s there—but because taste is expensive to lose. Sarah’s job is to draft, not to decide. The system generates a highly specific “Gary, let’s have lunch…” message with context: Gary’s last post was about his data pipeline, he complained about demo fatigue, he likes punchy emails, he attends Tuesday founder breakfasts, and he’s curious about AI that respects constraints. Great. Now pause. A human reads it. Maybe they remove one adjective, tweak the call-to-action, and make sure the opener doesn’t look like it was stitched from three scraped nouns and a wish. Then send.

Outbound isn’t binary. It’s a slope. Full automation feels efficient and then blows up your reputation in 30 days. Full manual feels artisanal and then you close 3 deals a year. The slope is where you live: templates evolve, guardrails tighten, the humans get smarter, and the system learns what “no cringe” means inside your specific niche. You are training your taste as much as your model.

From Prototype to Scale: The India Relay

The plan is simple: I prototype everything myself first. I wrap my hands around every edge—where it breaks, where it shines—and only then do I bring in dev partners from India to scale it out responsibly. Why India? I’ve got trusted partners there who understand iteration without ceremony. They won’t turn a draft into a monument. They’ll keep the work alive. That matters. Offshoring is not about cheap; it’s about clarity and cadence. If the prototype sings, they’ll keep it in key while we add instruments.

This is also where the data shape matters. Attio holds the golden record. Q’s holds the context. ConnectSafely drives the mechanical actions. SyncIn patches the human conversation into the graph. This means scaling is mostly about throughput and monitoring, not about rethinking architecture. The system’s soul survives the headcount.

Nostalgia Break: Airpad, Google Wave, and the Dream of Cooperative Software

Some of you remember Airpad. Some of you remember Google Wave. I remember how both felt: like we were five minutes away from a world where text, state, and presence fused into the same canvas. We didn’t get there then, but we’re inching back toward it now. Q’s feels like a modern echo of that dream—context becomes the application. Not “integration” in the checkbox sense, but the convergence of creation and reference. You’re not fetching from twelve tools; you’re operating inside a living memory palace.

The Sarah Factor isn’t a tool. It’s the ritual of maintaining that palace. Throw out yesterday’s furniture, keep the light, make room for the instrument you just found.

Credits, Clouds, and the Economics of Thought

We need to talk about cost models with grown-up honesty. AI has dragged software into a metered universe. The old SaaS fantasy—$29 a month for infinite value—is being replaced by usage pricing where every token, every call, every preview has a cost curve. It’s brutal for wandering minds and beautiful for precise builders. When hovering is a dollar a minute, you either get disciplined or you get a bill that eats your margin.

Strategies that help:

• Set session budgets. If the job isn’t shippable in 20 minutes, you’re not scoping; you’re browsing.

• Pre-bake context. Feed the system real artifacts—PDFs, strategy docs, examples—so it reasons once, not forever.

• Treat models like contractors. Assign them jobs with deliverables, not vibes with timelines.

• Use the cheapest competent model for the draft, the smartest model for the verdict.

  
  

A final point about model selection: your stack should never fear switching. If tomorrow Gemini rewrites the playbook for outbound persona nuance, switch. If next week GPT invents a visual that increases reply rates, switch. If Claude becomes a lettersmith on par with your top SDR, switch. Loyalty belongs to the outcome, not the logo.

What “Good” Looks Like in Outbound Right Now

Let’s be specific. A good system today:

• Pulls live context from LinkedIn, site content, and PDFs into a unified folder so your prompts aren’t hallucinating job titles.

• Generates two or three candidate messages per contact with structured rationales attached—why this hook, why this CTA, what signals we used.

• Inserts a 5-second review checkpoint for a human who knows the brand voice to approve or nudge.

• Logs everything to Attio with correct threading via SyncIn.

• Uses ConnectSafely for the actual send and follow-up schedule, keeping rate limits and account health sacred.

• Measures response quality, not just response rate, then updates the prompt patterns based on closes, not clicks.

  
  

Notice what’s missing: spray-and-pray. High-volume garbage is easy to automate. Taste is not. The Sarah Factor optimizes for taste at scale.

Why This Matters: The Founder’s Reputation Is a Non-Renewable Resource

When founders delegate outbound to a tool or a freelance automation stack without a philosophy, the brand pays. You get weird tone mismatches, creepy personalization, and a handful of meetings that feel like they were booked with borrowed trust. That’s not growth—that’s erosion. The whole point of Sarah/Signal is to protect the voice while amplifying the reach. If the system can’t sound like you on your best day, it doesn’t deserve your domain.

This is why we keep humans in the loop. Not because they’re faster, but because they remember the weight of a promise. AI can draft a message; only you know if you want to have the lunch you just proposed.

The Daily Practice

Every day, the same ritual:

• What broke? Blow it up.

• What felt heavy? Replace it.

• What was magical? Double it.

• What changed in the landscape? Try it.

  
  

Zoho out. LinkedHelper out. Attio in. SyncIn in. Replit for a sketch. Credits bite. Q’s for the build. ConnectSafely for the wire. Humans for the soul. India for the relay. Repeat.

This is how the system gets better—not with a roadmap, but with taste and nerve. You don’t protect the stack; you protect the principles.

Where We Are Now, What’s Next

Right now, Sarah’s generating and queueing outreach with full context in Q’s; logging and threading with Attio + SyncIn; shipping via ConnectSafely; and keeping me in the approval flow. The model choice is dynamic per job, and the cost profile is under control because we’ve tightened the session scope and stopped treating previews like free candy.

Next steps: round out the template library with real-world, closed-won examples; formalize the review UX so five seconds is truly five seconds; then bring in the India team to harden reliability and scale campaign throughput. I want speed without slop, automation without apology.

And yes, we’re still making “cool SDR experiences” at Signal. The goal is not just better emails—it’s a better feeling about the way we reach out. Because if the ops feel gross, the outcomes will, too.

Sponsors: Fuel for the Build

We need a sponsor for the show. If you want to back a build that’s allergic to fluff and obsessed with craft, let’s talk.

• Warm Up IP: If your cold email open rate is stuck at 10%, this is oxygen. Warm-up that actually moves the needle so your first impression isn’t a junk folder.

• Sorint: Builders of enterprise IT apps that don’t make users hate Mondays. If your legacy workflows are screaming for daylight, they’re your crew.

• 
  

Closing: The Joke, the Truth, the Work

The joke is credits. The truth is consumption. The work is taste.

I’ll keep eating popcorn on stream and forgetting to pay subscriptions at the worst possible moment. I’ll keep blowing it up and building it back. I’ll keep swapping models until the words feel like mine and the meetings feel like invitations, not ambushes. That’s The Sarah Factor: protect the voice, respect the data, and never let a tool tell you who you are.

If you want to see it live, you know where to find me. If you want to sponsor the chaos, even better. And if you’re still trying to make LinkedHelper behave like a brand—let it go. There’s better music on this side.