Ben Stolovitz, a research engineer at Salesforce AI Research, reported in October 2026 that he spends about $3,000 per month on AI tokens.reported He previously worked on Microsoft’s Windows AI team and noted significant changes in his workflow over the prior twelve months.He uses AI tools to manage coding tasks instead of writing the code himself.He manages multiple agentic workflows using tools like Claude Code, shifting from manual implementation to oversight.
What was released or claimed
Stolovitz published a blog post titled "How I use AI" detailing his shift from manual coding to managing autonomous agents. He described using tools like Claude Code, OpenCode, Codex, and Pi to handle tasks such as implementing features and reproducing bugs. Each agent operates on its own copy of the codebase and can write code, run tests, and create pull requests.
Who it affects
The post addresses software engineers and researchers who are integrating large language models into their daily workflows. Stolovitz noted that teammates are using AI to write Slack messages and that researchers are delegating thinking tasks to machines.
What is still unknown
Stolovitz stated that his personal AI policy is still in draft form and may be "utterly, hopelessly wrong." He did not specify which exact models account for the $3,000 monthly spend, only listing the agent names. He also noted that humans currently lack established norms for using these tools, leaving the long-term social impact unclear.
What to make of it
The author believes agentic coding has made manual coding obsolete for his specific role. He spends more time supporting scientists and building rapid prototypes than working with large engineering teams. He views the personal policy as a necessary framework to distinguish between helpful and harmful AI use as the technology becomes smarter.
What buyers should weigh
Organizations should consider the operational cost of agentic workflows, illustrated by the $3,000 monthly token spend reported by one engineer.reported Buyers must evaluate how agent-based development changes team structures, shifting focus from coding to agent management and prototype support. Vendor choice involves selecting from multiple agents like Claude Code or Codex, each requiring separate code copies. The lack of industry norms suggests companies may need to develop internal policies to guide usage and mitigate risks like over-reliance on automated thinking.
Why it matters
- High monthly token costs ($3,000) signal that agentic workflows may carry significant operational expenses for enterprises.
- The shift from coding to agent management redefines the skill set required for engineering roles in AI-heavy environments.
- The absence of established norms for AI use highlights a governance gap that policy readers must address in organizational frameworks.
- Multiple agent options (Claude, Codex, etc.) indicate a fragmented tooling landscape requiring careful vendor evaluation.
Topics: AI agents, coding tools, Salesforce, Microsoft, token costs
Updated
Then / now / next
The author wrote code manually and used GitHub Copilot for autocomplete assistance.
He manages multiple AI agents that handle coding, testing, and pull request creation.
He plans to adapt his personal AI policy over time as the technology evolves.
Story timeline
- Then The author wrote code manually and used GitHub Copilot for autocomplete assistance.
- 03 Nov 2026 How I use AI (Oct 2026)
- Now He manages multiple AI agents that handle coding, testing, and pull request creation.
- Next He plans to adapt his personal AI policy over time as the technology evolves.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
How much does the engineer spend on AI monthly?
The engineer reports spending approximately $3,000 per month on tokens for his AI usage.
What tasks are the AI agents performing?
The agents write code, run tests, create pull requests, and implement features or reproduce bugs.
Connected stories
Reading receipt
- Agentic coding tools have replaced manual coding for the author, who now manages multiple agents for features and bugs.
- The engineer is drafting a personal policy to establish norms for daily AI use, citing a lack of industry standards.
Still open: How will personal AI policies evolve into standardized industry norms?
Still open: What is the long-term impact on engineering teams when manual coding becomes obsolete?