What is Agentic AI - Autonomous Agents and Intelligent Workflows | CXOTalk #853

·56 min·5 clips

Praveen Akiraju explains why the term 'agentic AI' is weird and how it erupted from open-source projects like Baby AGI in 2023.

In CXOTalk episode 853, host Michael Krigsman welcomes Praveen Akiraju, a managing director at venture firm Insight Partners, to discuss agentic AI. Akiraju begins by addressing the term's sudden popularity, linking it to the release of open-source projects Autogen and Baby AGI in March 2023. He provides background on the current AI landscape, noting rapid advancements like OpenAI's O1 model and Google's cheaper models. Akiraju defines AI agents as applications that execute tasks autonomously, contrasting them with earlier automation technologies like RPA by emphasizing dynamic adaptation to user input and environments. He breaks down an agent's internal structure: a user interface with natural language processing, a planning and reasoning function, tool-calling or API integrations, a reflection loop for accuracy, and a governance policy layer. Akiraju highlights planning and reasoning as the critical barrier for true autonomy, citing techniques like chain-of-thought reasoning in models such as GPT-4o1. Listener questions drive much of the discussion, covering topics like enterprise back-office automation, the composition of agents (mixing LLMs with traditional machine learning models), and job displacement concerns. Akiraju argues that AI agents will automate repetitive tasks but enhance productivity for roles like developers and analysts, comparing current AI to a 'competent intern' and future agents to a 'competent colleague.' He stresses trust as foundational, involving accuracy and safety guardrails. On implementation, Akiraju describes building agents as an engineering integration problem, recommending data infrastructure readiness and mentioning companies like Crew AI, Workato, and Salesforce's AgentForce. He discusses data strategies such as fine-tuning and RAG for domain-specific use. The episode concludes with thoughts on hype, deterministic outcomes from non-deterministic LLMs, and governance to prevent agents from 'going off the rails.'

As heard by us

Agentic AI is framed as a workflow and governance problem, not a slogan.

CXOTalk treats agentic AI as a practical business problem, asking what changes when software can absorb repetitive work, coordinate workflows, and push enterprises toward clearer governance.

Read the full review in PlayNext →

Why you'd press play

You want agentic AI explained in terms your calendar, team, and budget can use. This episode frames agents and intelligent workflows as practical tools, not magic, and keeps coming back to enterprise reality.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from CXOTalk