Tubegraph

Agent Engineering and its Definition

4 min read

From: aidotengineer

The Evolution of AI Engineering

The field of AI engineering has been undergoing significant maturation and expansion across various disciplines [01:19:00]. Previous conferences have landmarked its progress:

  • “The Rise of the AI Engineer”: Focused on the emergence of the AI engineer [01:11:00].
  • “The Three Types of AI Engineer”: Explored different archetypes within the field [01:13:00].
  • “AI Engineer World’s Fair”: Highlighted the discipline’s maturity and spread [01:17:00].

While many initially viewed AI engineering as either an extension of machine learning (ML) with some prompt engineering [01:41:00], or primarily software engineering with calls to large language model (LLM) APIs [01:47:00], it is emerging as its own distinct discipline [01:55:00]. It is currently often described as “90% software engineering, 10% AI,” but this AI component is expected to grow significantly over time [02:03:00].

Defining Agent Engineering

The AI Engineer Summit has pivoted to become an Agent Engineering conference [02:44:00]. This shift signifies a focused approach, opting to exclude topics like RAG, open models, and GPUs in favor of concentrating solely on agents [02:53:00]. This decision was influenced by YouTube’s top-performing talks, which indicated a strong audience interest in “agentic things” [03:18:00]. A new rule was implemented to prioritize talks from those using agent frameworks in production rather than just vendors creating them, making content curation more challenging but valuable [03:32:00].

A common phrase, “2025 is the year of Agents,” is a prediction pushed by prominent figures like Satya Nadella, Roman, Greg Brockman, and Sam Altman [04:20:00]. While there was initial skepticism among conference organizers, including a past recommendation to remove “agents” from branding [04:48:00], that advice has now reversed [05:17:00].

What is an Agent?

Defining an agent is a “monumental task” [05:33:00]. Different perspectives exist:

  • Machine Learning (ML) perspective: Focuses on reinforcement learning environments, actions, and achieving goals [05:41:00].
  • Software Engineering (SE) perspective: Tends to be more reductive, often simplifying it to a “for loop” [05:49:00].

Simon Willison, considered a “patron saint” in the AI engineering community, has crowdsourced over 300 definitions for what constitutes an agent [06:01:00]. Common themes among these definitions include:

OpenAI also recently released a new definition for agents, which they are building upon [06:52:00].

Why Agents are Working Now

The current success of agents, compared to a year or two ago, can be attributed to several factors:

  • Increased Capabilities: Agent capabilities are rapidly growing and are beginning to hit human baselines [07:19:00].
  • Improved Reasoning and Tool Use: Models now possess better reasoning and tool-use abilities [07:37:00].
  • Model Diversity: The market share for models like OpenAI has diversified, with OpenAI’s share dropping from 95% to 50% in two years, and new Frontier Model Labs emerging as potential challengers [07:51:00].
  • Lower Cost of Intelligence: The cost of GPT-4 level intelligence has decreased by 1,000 times in the last 18 months [08:14:00].
  • RL Fine-tuning Options: New options for reinforcement learning (RL) fine-tuning are becoming available [08:28:00].
  • Charging for Outcomes: The shift towards charging for outcomes rather than just costs [08:43:00].
  • Multi-agents and Faster Inference: Advances in multi-agent systems and faster inference due to improved hardware [08:49:00].

Agent Use Cases

Several agent use cases have demonstrated product-market fit (PMF) [09:06:00]:

However, certain “anti-use cases” are discouraged from being demonstrated, such as agents that book flights or Instacart orders, as users often prefer to retain control over these tasks [09:24:00].

Impact on AI Product Growth

The growth of AI products, particularly platforms like ChatGPT, is tightly linked to reasoning capabilities and the deployment of agentic models [10:39:00]. OpenAI reported 400 million users, a 33% growth in three months [09:46:00]. ChatGPT experienced a period of stagnant growth until it began shipping agentic models [10:09:00]. The introduction of 01 models doubled ChatGPT’s usage [10:24:00], projecting it to reach one billion users by the end of the current year, a fivefold increase from September of last year [10:28:00]. This growth indicates a significant opportunity for the wider AI industry [10:53:00].

The Future of AI Engineering

The role of the AI engineer is evolving to focus on building agents, similar to how ML engineers build models and software engineers build software [11:00:00]. This shift underscores the increasing importance of agent engineering as a core discipline within the broader AI landscape.

Graph View

Backlinks