What are the key challenges organizations face when adopting AI agents, and why?

The answers highlight diverse challenges—ranging from governance, data quality, control, cost, and technical debt to infrastructure, agent management, and employee burden—reflecting little consensus but broad concern across technical and organizational fronts.

12 answers

Answers

  • Paul Roetzer

    Founder & CEO · Smarterx

    Risk and Governance

    Paul Roetzer says organizations should first understand their existing limitations and the risk and governance concerns of stakeholders, while not letting restrictions block transformation by finding low- or no-risk use cases.

    The answer starts at 6:30.

    The Artificial Intelligence Show · Listen on Apple Podcasts
  • Stijn (Stan) Christiaens

    Co-founder & Chief Data Citizen · Collibra

    Trustworthy Data

    Stijn (Stan) Christiaens says trustworthy data is a key challenge organizations face in adopting AI agents, noting that data must be trustworthy before it is useful.

    Bernard Marr · Watch on YouTube
  • Amjad Masad

    CEO · Replit

    Harder to Control, Cause Subtle Bugs

    Amjad Masad says companies adopting AI agents will find them harder to control and that they introduce bugs in new, subtle ways.

    TechCrunch · Watch on YouTube
  • Massimiliano Marcon

    Director, Product Management, Atlas Agent Engine · Mongodb

    High Cost and Rapid Change

    Massimiliano Marcon says the challenges are high cost and rapid change: customers who dismissed cost later returned saying it was a problem, and framework changes can eat six months of rebuilding while models and frameworks move fast.

    Dev Interrupted · Dev Interrupted · Read
  • Rama Akkiraju

    VP, Enterprise AI · Nvidia

    Technical Debt

    Rama Akkiraju says technical debt is a challenge organizations face with AI agents, because it is immensely magnified, can wear you down, and can affect resourcing strategy when moving fast on AI.

    Imagination in Action · Watch on YouTube
  • Florian Douetteau

    CEO at Dataiku · Dataiku

    Agent Management

    Florian Douetteau says the challenge is agent management: companies don't even know how many agents they run, and he argues they need a control plane tracking each agent's cost, value, behaviors, risks and ownership — inventing agent management as they go.

    The answer starts at 25:24.

    The Deep View: Conversations · Listen on Apple Podcasts
  • Ben Lorica 罗瑞卡

    Principal · Gradient Flow

    Lack of Decision‑Making Layer

    Ben Lorica 罗瑞卡 says the key challenge is the lack of a decision-making reasoning layer: LLMs can retrieve but cannot make decisions, so agents need a symbolic model of cause and effect and prioritization.

    The answer starts at 16:01.

    The Data Exchange with Ben Lorica · Listen on Apple Podcasts
  • Arvind Jain

    Founder and CEO · Glean

    Employee Overload

    Arvind Jain says employees spend a lot of time coaching AI agents and feel overburdened by it, calling it "bot sitting" — working for AI rather than AI working for them.

    Product Market Fit · Product Market Fit · Read
  • Roberto Milev

    Chief Architect · Navan

    Cost Predictability

    Roberto Milev says a key challenge is cost predictability, because it is very hard to predict and manage agent costs and add guardrails, with vendors' incentive to spend more tokens.

    AI Engineer · Watch on YouTube
  • Angela Jiang

    Head of Product, Claude Platform · Anthropic

    Infrastructure Demands

    Angela Jiang says organizations need a lot of infrastructure, including ways for agents to recover from errors and work securely and compliantly, which she calls a different class of problems.

    Kleiner Perkins · Watch on YouTube
  • Rob Allen

    Chief Product Officer · Threatlocker

    Weak Guardrails

    Rob Allen says the built-in guardrails for AI agents are pretty flimsy, and that most users, including advanced ones, don't understand what these tools are doing underneath.

    Business Security Weekly (Video) · Business Security Weekly (Video) · Listen at 6:35
  • Pete Johnson

    Field CTO, Artificial Intelligence · MongoDB

    Picking Wrong Problems

    Pete Johnson says the challenge is organizations picking the wrong problems, since they need problems with good data and measurable success metrics.

    Super Data Science: ML & AI Podcast with Jon Krohn · Watch on YouTube