Claude Runs Ads. The Hard Part is Operations at Scale.

Before I founded Synter, I spent a year at Sourcegraph where I got early access to Amp, their AI coding agent, before it launched publicly. I connected Amp to our Google Ads account and started testing it publicly in August 2025.

Over the next several months, I ran significant ad spend with AI agents across LinkedIn, Google, Reddit, X, Microsoft, and StackAdapt. I've been running ads with AI agents for nearly 9 months now. I've learned a lot. And I'm still not fully an expert. That should tell you something about the depth of this problem.

This post is the story of what I learned, and why I ended up building Synter instead of maintaining a folder of scripts.

Open Claude, Cursor, or Amp. Ask for a Google Ads API script. Maybe an OAuth flow. A lightweight reporting query. Maybe a bid adjustment worker that runs on a cron.

It feels rational because it is rational. You learn fast. You stay close to the logic. You feel in control. Early wins come quickly.

If you're technical, building your own ad tooling is almost irresistible. And for the first few days, it feels like a superpower. I know because I lived it.

The first script is not the product. The product is everything required to keep it running across real accounts, real spend, and real platform changes.

What starts as “just a script” becomes:

• Auth handling and token lifecycle management
• Job scheduling with retries and backoff
• Logging, alerting, and error triage
• Reporting normalization across platforms
• API version migrations when platforms deprecate endpoints
• Edge-case cleanup you didn't know existed

Claude can get you to working code quickly. It cannot make platform churn disappear.

At Sourcegraph, I had early access to the best AI coding tools available. I used them aggressively. And I still burned weeks on platform-specific quirks that no amount of prompting could anticipate.

Every ad platform has its own flavor of complexity. Here's what maintaining integrations across 8+ platforms actually looks like, from someone who has built and maintained all of them:

Every network names the same concept differently. Google has campaigns with ad groups. LinkedIn has campaigns with campaign groups. Reddit has campaigns with ad groups but different bid strategies. Amazon DSP renamed orders to campaigns and line items to ad groups in v1.

Attribution windows don't match. Timezones don't match. Reporting schemas don't match. Creative review pipelines don't match. Currency handling differs. Rate limits differ.

The issue isn't whether you can write an integration. It's whether you want to become the person maintaining eight of them forever.

Once you automate media buying, the next bottleneck is creative throughput. You need variants. You need platform-specific formats. You need to test and iterate faster than your design team can produce.

Builders who start with “I'll automate Google Ads” end up needing asset generation, copy iteration, video workflows, structured naming, and feedback loops from performance back into creative production.

You're not building a script anymore. You're building a system.

People ask me: “Why is Synter better than Claude Code?” That's the wrong question. Synter uses Opus 4.6 under the hood too. The model is not the differentiator. What matters is the harness you build around it. That's what I learned during my time at Amp.

The question is whether you want to maintain your own ad ops machine or use one that's already built. For a lot of people, Synter is the easier path.

The harness Synter provides:

• 100+ scripts across 8+ ad platforms with platform-specific API quirks resolved
• OAuth and token lifecycle for every platform, including refresh tokens, re-auth detection, and connection health monitoring
• Campaign execution through natural language. Create, pause, update budgets, manage creatives.
• Creative Studio for generating images (Imagen 4, Flux) and video ads (Veo 3.1, Runway, Luma, Creatify) from a prompt, with brand guidelines auto-applied
• Server-side conversion tracking that fires to Meta, Google, Reddit, LinkedIn, X, TikTok, and Microsoft in parallel

Synter also connects not just to ad platforms, but to your CRM, Google Tag Manager, conversion infrastructure, and reporting stack. That means less setup, less breakage, faster execution, and a system that compounds learnings over time.

This isn't an anti-builder argument. I am a builder. There are smart things to build yourself:

• Custom internal reporting and attribution models
• Proprietary scoring and lead qualification logic
• CRM enrichment and post-conversion workflows
• Analytics layers specific to your business

But the lowest-leverage place to spend engineering time is re-implementing OAuth, chasing API changes, and normalizing eight platforms. Build where your business is differentiated. Don't rebuild the commodity chaos.

Use Claude to build what's unique to your business. Don't use your best operators and engineers to babysit refresh tokens.

Here's the thing: you don't have to choose. Synter ships an MCP server that plugs directly into Claude, Cursor, Codex, Amp, and ChatGPT. You keep using your favorite AI coding tool. Synter handles the ad platform complexity behind it.

100+ tools. 8+ platforms. OAuth, credentials, and conversion tracking handled for you. One line to install.

People ask if Synter can outperform their agency. Maybe. But I think the better test is Claude out of the box versus Synter out of the box.

Synter is not autonomous yet. The brand is still in control of their campaigns. Our focus is not on outperforming Claude or GPT-5.4 or Gemini Pro. Our focus is on removing the hassle of managing API keys, re-authentication cron jobs for your tokens, creatives, collaboration with your team, and more.

The lesson from 9 months of running ads with AI agents: the first script is free. The second platform is annoying. The third is a full-time job. By the eighth, you've built an ad ops team whether you meant to or not.

Synter exists so you don't have to build all of that yourself.

— Joel