Have your AI agent log its own API calls for a week before vendors start charging for agent access, then cut calls or mirror data to Postgres
Facing an estimate of up to $240,000 a year as SaaS vendors begin charging for agent API access, SaaStr had its main revenue agent track its own calls for a week. It found calls to cut and proposed a $5 Postgres mirror of the system of record.
Evidence: The author reports this. We have not checked it beyond reading the source.
The business problem
An agent that runs revenue operations makes tens of thousands of API calls a day into CRM and other vendor apps. As vendors such as Salesforce, HubSpot and Atlassian add charges or limits for agent access, those calls could become a large and unplanned cost.
What was tried
1. Ask the agent (SaaStr's in-house agent, called 10K) to install a tracker on its own API usage and run it for a week, so you see which vendor calls it makes and how often. 2. Have the agent review the log and flag calls that can be removed or reduced as a short-term fix. 3. For vendors whose pricing could become expensive, mirror the system of record into a cheap Postgres database (Supabase was mentioned) and point the agent's reads at the mirror instead of the vendor API. The agent offered to set this up the same day. 4. Accept the trade-off: two systems must stay in sync and will sometimes drift. The same episode notes that a vendor that once capped agent API access to 10 to 20 minutes a day led SaaStr to move off it.
What was reported (not reported)
The agent reportedly found many calls that could be cut, and a mirror is planned for any vendor whose pricing nears the estimate. No numbers are given for calls removed, money saved, or whether the mirror was built. The $240,000 figure is a single estimate, and vendor pricing was not yet known.
Limitations
SaaStr is a media and events company with three people and more than 20 agents making 35,000 to 40,000 API calls a day, far more than a typical solo founder. No code, tracker design or before-and-after figures are shown. The post is a write-up of a podcast episode on SaaStr's own site with promotion for its events and playbook, and much of the page covers unrelated topics (attendee data, ad tests). The publication date is not stated on the page; it appears to be from late October 2026.
What you need
An agent that calls vendor APIs on your behalf, and a way for it to log its own requests.
Sources
- SaaStr (Jason Lemkin) ↗ Firsthand write-up, publication date unknown
Source published: unknown. Last reviewed here: November 2, 2026. Spot a mistake? Tell us.
Tools in this workflow
- Supabase: Hosted Postgres option mentionedTry Supabase
- HubSpot: Vendor app the agent callsTry HubSpot
- Salesforce: Vendor app the agent callsTry Salesforce
- PostgreSQL: Low-cost mirror of the system of recordTry PostgreSQL
Get new workflows like this by email
A short selection of the most useful new workflows, including the ones that did not work. Free, and you can unsubscribe at any time.
Related workflows
Audit 38 fresh Show HN launches for AI-assistant readiness: eight page checks plus asking a model if it knows the product
A one-person SaaS founder pulled a month of Show HN launches through the Hacker News API, ran eight technical checks on the top 38 sites and asked one language model whether it knew each product. It could name 3.
Rank what to build next with an open-source MCP server that scores support tickets and feature requests, then lets Claude write the plan
A product lead built an MIT-licensed MCP server that pulls Help Scout tickets and ProductLift requests, strips personal data, groups them into themes, scores them with a weighted formula and hands the result to Claude, with business KPIs able to reorder the list.
Deliver adaptive micro-courses in Telegram with GPT-generated lessons and Supabase learning state
The builder of SkillDay describes an n8n learning bot that generates courses, quizzes and feedback with GPT-4o-mini, stores learner state in Supabase and adjusts review intervals from test results.
When an AI marketing agent's test and your own test change the same clips on the same day
In a public log of AI agents selling small products, a marketing agent proposed a sound test of why TikTok likes fell. The operator found the data could not answer it, because their own music test had started on the same day on the same clips.