MEEcom

I build ecommerce systems that scale. Paid ads. UGC. AI leverage. If you care about execution > motivation, you’ll want to follow. Lock in.

Found a Google tool most ecom founders haven't touched yet. It's changing how I approach creative testing.

Someone I follow mapped their entire outreach stack last week. 20 n8n workflows. Cold email, lead enrichment, AI calling, CRM sync. What their manual team used to cost: $10-20k/month. Hard to unsee that math once you see it.

LinkedIn connection accept rates were under 15% for months. We tried everything. Better copy. Targeted filters. Even paid for Sales Navigator. Nothing moved the needle until we stopped typing and started recording.

Dropped $300/video UGC for Arcads last month. Same brief. Same offer. AI avatar on the hook. Human version: 3 weeks of revisions. AI version: live same afternoon. ROAS held. I still don't know how to feel about that.

Been in ecom long enough to watch three "everything will change" waves. AI tools are the first one where I'm actually changing how I work. But there's a ceiling hitting faster than anyone admits.

$2.5M/mo ceiling. 70 coaches, $25K/day in ads. That's not a bad business. It's just not scalable. You're still selling time, AI just makes it cheaper. Kajabi hit $100M+ ARR. Skool, Whop, same path. All started as services and got tired of the ceiling. Services teach you what to build. Then you build the infrastructure under yourself. Who's making that jump right now?

The less you need a UI, the more leverage you get.

LangGraph vs a 200-line Python scheduler calling `claude -p`. LangGraph: elegant state graphs, retry logic, graph visualization. Simple orchestrator: one file, 30 agents on a schedule, each with its own prompt file. Built the simple one over a weekend. LangGraph is still sitting in a branch.

@TheecomMike That flat repeat AOV is almost always a product discovery problem. Returning customers have higher intent but get shown the same homepage as strangers.

i rebuilt a post purchase flow last october. repeat AOV went from $79 to $118 in 10 weeks. same store. same customers. same products. here's what i found when i audited it: first order AOV: $76. repeat order AOV: $79. almost identical. that gap told me everything. customers coming back were buying the same thing they bought the first time. nobody was moving up to bundles. nobody was buying the higher ticket products. nobody was being shown what to buy next. then i looked at the post purchase flow. thank you email with a discount code. shipping update. review request at day 14. the discount code in the thank you email was actually making it worse. it was training repeat customers to expect a discount before they'd buy again. compressing margins on the customers with the highest LTV potential. and still not moving anyone toward higher ticket products because the email had zero product direction. so i rebuilt the whole sequence. removed the blanket discount from the thank you email entirely. built a 4 email sequence that educated customers on the full product line based on what they bought first. made specific product recommendations timed to when cohort data showed customers were actually ready to buy again. moved discounts to the winback sequence only for customers who hadn't purchased in 75 days. repeat AOV went from $79 to $118 in 10 weeks. a blanket discount in your thank you email isn't a retention strategy. it's margin compression dressed up as one.

@TheecomMike That gap is the tell. When repeat AOV matches first-order AOV, the post-purchase sequence is doing nothing but confirming the sale. What did you change first, the timing or the offer?

@anKit0017_ @trikcode The code is just infrastructure. The real bottleneck is whether the human running the AI knows what output they're actually trying to get.

@trikcode Ppl worried about AI being too good usually forget what's really scary: humans still writing the code

Cloudflare stock dropped 13% because of an AI model that hasn't even been released yet. Let that sink in. Not a data breach. Not even a hack. Anthropic built something so dangerous that it moves markets The threat isn't AI being dumb anymore. It's AI being too good.

Ecom AI tools running on stale models and nobody notices. Your chatbot is giving customers last year's answers.

@artyomvnsv @trikcode Both can be true. Hype cycles are real, but the brands actually using AI to cut creative production time are keeping those gains whether or not the valuations hold.

@trikcode Either that or we're approaching peak hype and markets are behaving irrationally

Claude 3.7 Sonnet for positioning briefs and brand voice work. GPT-4o for structured JSON extraction from review data. Tried swapping them. Claude over-explains the schema. GPT flattens the nuance. Wrong model in the wrong slot costs more than the API bill.

Spent $300k+ on ads. The product was the problem every single time.

@aaditsh that's why I stopped evaluating tools on feature lists. the question is whether it actually changes how my team moves through work day-to-day.

Anthropic hitting $30B ARR is incredible. But that's not the actual story. The story is the decisions they made 18 months ago that nobody was paying attention to. Quick rundown of what they actually shipped: 1. Claude Code. Not autocomplete. Delegation. You hand it a whole repo and it runs the project end to end. I was at Anthropic HQ in SF for the launch day. They invited me out for it. The room felt different. Everything they said that day about what Claude Code would do has played out. From zero to $2.5B run rate in 9 months. Faster than any enterprise software product in history. 2. MCP. Open-sourced November 2024. Basically USB-C for AI agents. Build the integration once and Claude talks to your whole stack. 5,800 servers now. 97M monthly SDK downloads. Once a company runs its systems on MCP, leaving means rebuilding everything from scratch. That's a moat most people don't even register as a moat. 3. Available on AWS, Google, and Azure. The only frontier model on all three. If you're already an AWS customer, turning on Claude is one contract amendment. That single decision shortened the enterprise sales cycle more than any feature launch. 4. Max plan at $200 flat. Sounds boring. It's not. Per-token pricing makes heavy users ration. Flat pricing makes them run Claude constantly. Constant usage means Claude is inside the workflow. Inside the workflow means switching costs compound every month. They used pricing to hardcode themselves into developers' daily lives. 5. EU AI Act pre-compliance. Banks, hospitals, insurers could approve Claude deployments faster because legal already had what they needed. Boring, important. And the models are actually best in class. Sonnet and Opus are the ones I reach for first when I'm doing real work. Two years ago, 12 companies were spending over $1M a year with Anthropic. Today it's over 1,000. That number doubled in two months. Anthropic now wins 70% of head-to-head enterprise deals among new AI buyers per Ramp's March 2026 AI Index. The thing that could still break this: compute. Anthropic is turning away revenue because they can't serve it. The Broadcom + Google TPU deal they just announced is them trying to fix it before it becomes the story. One last data point I'll share. I run an AI newsletter with 100K readers. The questions in my inbox flipped this year. A year ago people were asking which custom GPTs to use for their business. Now they're asking which Claude plan to upgrade to. That shift in inbound is the leading indicator I trust most because I see it first-hand on a weekly basis. Anthropic didn't win in April 2026. They won in November 2024. April is just when everyone noticed.

@aaditsh that's the trap most evals miss. You can benchmark model quality all day, but if the tool doesn't fit the actual motion of work, people route around it.

@sam__works GPT copy always sounds like a press release. Claude actually understands why a customer hesitates, which makes all the difference for PDPs and email flows.

@MEEcom44 Claude for nuance and judgment is the right call. I tried GPT for copy and it writes like a robot every time.

Built an ad creative pipeline and ended up with three models doing different jobs. Claude handles copy and judgment calls. Anything that needs nuance, brand voice, or a decision. It's not the fastest but it's the one I trust. Veo 3 for video with native audio. No voiceover budget. No creator brief. Just a product and a prompt. Llama for high-volume repeatable stuff. Product descriptions, tag generation, email subject line variants. Cost per call matters when you're running thousands of them. One model for everything is like using the same ad creative for cold and retention. Works until it doesn't.

@sam__works GPT copy always feels like it's trying to sound smart. Claude actually sounds like a person, which matters a lot when you're writing product pages or emails where trust is everything.

@_SilasVane Exactly, the asset library plus a tagging system for what actually converted is where it compounds. Most teams shoot, post, forget, and wonder why their CPC keeps climbing.

@MEEcom44 The speed jump is real. The interesting part now is who builds review + reuse into the workflow instead of treating every clip like a one-off.

AI video tools just made me rethink our entire creative pipeline. What used to take a week of UGC coordination now takes an afternoon. We're not there yet but the gap is closing fast.