Krish Ramineni is CEO & Co-founder at Fireflies.ai, an AI teammate for meetings used by people at 75% of Fortune 500 companies.
There are millions of articles about how to be a good people manager. Harvard Business Review alone publishes thousands of think pieces annually on leadership, emotional intelligence and team dynamics. But as AI rapidly integrates into daily work, we’re missing a crucial conversation: How do you effectively manage AI agents?
The shift in perspective matters. When you think of AI as just another “tool” to use, you’ll never unlock its full potential. But when you think of AI as a teammate to manage, everything changes. Suddenly, the management skills that made you successful with humans become the foundation for multiplying your impact through AI.
As I wrote recently in Fast Company, middle management is dead, and every employee is becoming a mini-CEO. The key skill these mini-CEOs need? Exceptional AI management.
Right now, we’re teaching everyone the basics: how to write a single prompt to get AI to complete a single task. But managing AI means orchestrating multiple agents simultaneously, evaluating their outputs, iterating rapidly and building systematic workflows that actually scale.
That’s like teaching someone to use a calculator but never teaching them accounting. Sure, they can add numbers. But can they run the books for an entire company?
What Stays The Same
The fundamentals of good management translate surprisingly well to AI. Clear communication remains paramount; vague instructions produce poor results, whether you’re directing humans or machines.
Training and onboarding matter just as much. You have to teach and train your AI like you would onboard a new employee. The best AI managers provide context and examples, then iterate based on results.
“Trust but verify” applies equally. Whether managing people or AI, you need systems to check quality and catch errors before they impact customers. And just like building diverse human teams, you need various AI agents. Your writing AI won’t excel at data analysis, just as your star salesperson might struggle with spreadsheets.
What Changes Completely
Here’s where AI management becomes truly fascinating: First, the feedback loop is instantaneous. With humans, you wait weeks or months to know if your management approach works. With AI, you know in seconds, which enables rapid learning and improvement that’s simply impossible with traditional management.
My company, Fireflies, recently held a “hackathon.” The team built AI agents for everything from shipping design components to auto-generating social media launch posts. A first attempt at the launch post produced generic, corporate-sounding content. But within minutes, the team iterated through five versions by adding brand voice examples, specifying our tone and including product context. Soon, the AI was writing posts indistinguishable from our best human copywriters.
Second, the scale multiplies dramatically. Traditional management theory suggests a span of control of seven or eight people. The most productive AI managers handle 10 to 15 agents simultaneously, a productivity multiplier that transforms what one can accomplish.
Third, there’s the “throwaway reality”—it’s normal to throw away half of your AI work as you learn and iterate. This is the new normal; the skill lies in rapid evaluation and course correction.
The Three Skills That Matter Most
As CEO of an AI-native company and as someone who’s spent plenty of time learning to manage AI, there are three skills I’ve seen drive the most impact:
1. Prompt Iteration Mastery
The best AI managers don’t write perfect prompts—they iterate. Start broad, analyze the output, then narrow with surgical precision.
Your first prompt might be: “Analyze our customer feedback from last quarter.” The AI returns a wall of text. Too broad.
Second attempt: “Identify the top 3 customer complaints from Q4 2024, categorized by feature.” Better.
Third iteration: “Analyze Q4 2024 customer feedback. Group complaints into categories: UI/UX, Performance, Pricing, Features. Show count and percentage for each. Include one representative quote per category.” Now you’ve got actionable insights.
Each iteration takes seconds, not days. In the time it takes to schedule one meeting, you’ve refined your approach five times.
2. Systematizing What Works
Stop writing prompts from scratch. Build reusable frameworks, templates and examples that improve AI performance over time. The compound effect is staggering.
Take a sales manager who needs weekly pipeline reports. Using prompt iteration mastery, they quickly determine what works to create the perfect prompt. This takes five minutes.
Here’s where systematizing comes in. They don’t just use this prompt once. They:
• Save it as a template
• Add example output showing exact formatting
• Note which data sources to reference
• Include definitions for company-specific terms
Next week? The report takes 30 seconds. By the third month, their entire sales team is using these templates. What started as a five-minute experiment becomes a system that saves hundreds of hours.
At Fireflies, we’ve turned this concept into reality through automated workflows, for both customers and team members, that chain together to create systems. Our sales team’s pipeline reports are generated from call data automatically every Monday. Customer success gets real-time churn alerts based on sentiment analysis. Product teams receive auto-categorized feature requests. The compound effect? What started as individual time-savers evolved into an interconnected AI system that understands our business deeply and operates at 100 times the speed.
3. Parallel Processing
Forward-thinking leaders are building their own project management systems to help manage multiple AI agents. Instead of sequential tasks, your marketing campaign may involve coordinating three agents with their own unique assignments:
• Agent 1: Analyzing competitor campaigns for insights
• Agent 2: Generating social media content variations
• Agent 3: Creating email sequences for different segments
Each agent works simultaneously, checking in at different intervals. Your job is to review, redirect and synthesize. When Agent 2’s social content misses the brand voice, you provide examples and let it retry. When Agent 1 surfaces a brilliant competitor strategy, you feed that insight to Agents 2 and 3.
The mental shift is profound: You’re not a taskmaster assigning work sequentially. Instead, you’re a conductor ensuring every section of your AI orchestra plays in harmony.
Your Path Forward
Those who master AI management now will have an enormous advantage. Start small by picking one workflow, deploying an AI agent and practicing these skills. Iterate relentlessly. The future belongs to those who can multiply their impact through AI orchestration.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
