The claim: We built AI agents to replace HubSpot
Someone on X said they built AI agents that replaced everything HubSpot did for them, and they are not technical. Jason Lemkin believes them. SaaStr did something similar: custom AI agents handling SDR work, marketing ops, and customer success for 100+ sponsor relationships. Three humans, 20+ agents in production, closing real revenue.
But SaaStr did not replace Salesforce. They went headless on top of it.
Salesforce is still the system of record for every deal, contact, and ticket. The agents read and write to it via API. Almost no human logs in anymore. That is the actual move: buy the platform, build the last mile on top. Not vibe code your own CRM from scratch.
Here is why that distinction matters for sales orgs.
1. It breaks at scale
A solo founder fixing their own agent logic works. Put 40 AEs on it, then 400, and you need permissions, territory rules, audit logs, onboarding that does not require the founder on Slack. CRMs became a category because they imposed shared structure on lots of people who would otherwise do it their own way. Agents make it faster to build the thing. They do not make 400 reps agree on a sales process.
2. Integration hell is real
B2B revenue data lives in 30 places: billing, product analytics, marketing automation, ticketing, e-signature, the accounting system finance refuses to switch. Salesforce and HubSpot already did the miserable work of connecting to all of that and keeping it alive when vendors change APIs with no warning. SaaStr graded 116 B2B APIs this year. The quality spread is enormous. When you roll your own, every integration becomes your problem. Forever.
3. Maintenance never ends
Building it is 10%. Keeping it running is 90%. Models change. APIs change. Your business changes. SaaStr's main marketing agent is 14,000+ lines of code across 74 files and 373 commits. Their customer success agent manages 100+ live relationships, so downtime is not abstract. It works because someone owns it and tends it constantly. Most "I built it and I'm not technical" stories mean "I built it, it works today, and I haven't hit the week where it breaks and I can't fix it."
The actual wedge for AI agents
SaaStr is not arguing against agents. They run on them. But the pattern that works is: platform as data layer, agents as interface, humans as executors. Salesforce just launched headless for AI agents. SaaStr has been living it for six months. That is where this is going: not replacing the CRM, but never logging into it again.
For sales teams, that means: your CRM budget is not going away. Your workflow layer is. The platforms know this. Salesforce launching headless is them saying the quiet part loud. Pay us for the system of record, build whatever you want on top. Most orgs will take that deal because the alternative is owning the integration and maintenance burden themselves.
The pitch that you can vibe code your own HubSpot is technically true and strategically wrong for most teams past 10 people. The real move is building the agent layer on top of the platforms that already handle the boring, mission-critical stuff. That is what SaaStr did. That is what works.