Agent Engineering Status Report Overview

# Entry

LangChainone of today’s leading platforms for building and orchestrating artificial intelligence (AI) applications based on enormous language models (LLM) and agent engineering, recently released the platform Agent Engineering Status Reportwhich surveyed 1,300 professionals with different roles and business backgrounds to understand the current state of this noticeable AI trend.

This article highlights some of the report’s most significant findings and insights and discusses them in a more accessible tone, revealing some of the key terms and jargon related to AI agents. For more information on the key concepts of AI agents, please see the related article.

Before we focus on the facts, figures and evidence supporting each of our three carefully selected observations, here are some key terms and definitions you should know, briefly explained:

# Huge enterprises are overtaking start-ups in production

Key concepts worth knowing:

• Agent: An artificial intelligence system that, unlike standard chat-based applications that react reactively to user interactions, is able to make decisions and actions on its own. In the most commonly used context today, agents apply the LLM as their “brain,” helping them decide what next steps to take—for example, searching a database, sending an email, or searching the Internet—to achieve a goal.
• Production (environment): Although this is a basic software engineering concept, it may seem unfamiliar to readers from other backgrounds. Being “in production” means that the software system is live and real users, customers or employees are using it to perform some work or activity. Essentially, it is what follows a prototype or proof of concept (PoC): a test version of software that has been run in a controlled environment to identify and fix possible problems.

The most significant facts in the report:

• While there is a common bureaucratic misconception that larger companies are slower to adopt recent technologies, the data shows otherwise: they are leading the way in adopting AI agents – 67% of organizations with more than 10,000 employees have put agent-based applications into production, while only 50% of smaller organizations with fewer than 100 employees have done so.
• The reason for the above point may be the cost of building reliable agent solutions, with the need for significant investments in infrastructure.

Similar evidence can be found, among others, The state of artificial intelligence in the Deloitte enterprise in 2026 AND McKinsey’s State of Artificial Intelligence in 2025 reports.

# The observability-assessment gap

Key concepts worth knowing:

• Observability: Artificial intelligence models, especially advanced ones, are often perceived as unclear “black boxes” with unpredictable results. Observability is the ability to examine and record what the AI ​​“thinks” and how this leads to decisions or outcomes.
• Tracing: a specific aspect of observability, which involves recording the step-by-step journey of an AI agent – ​​i.e. its reasoning path.
• Offline rating: involves looking at a test dataset with known “correct” answers to measure how accurately and effectively an AI agent (or other AI system) performs.

The most significant facts in the report:

• An astonishing 89% of respondents from all backgrounds have implemented observability, although only 52.4% conduct offline assessments, revealing a noticeable disconnect between how teams monitor AI agents and how rigorously they test their performance.
• This signals a “ship and watch” mentality in which engineering teams prioritize debugging errors after they occur rather than preventing them before being deployed to production. Repairing “broken robots” rather than ensuring they function properly before leaving the “factory” can have undesirable consequences and costs.

Similar evidence can be found, among others, Observability of LLM Giskard and assessment article.

# Cost is no longer the main bottleneck: quality is

Key concepts worth knowing:

• Hallucinations: When an AI model such as LLM confidently generates false or nonsensical information as if it were real, it is said to be hallucinating. This is a hazardous problem when AI agents enter the loop, because the problem is not just saying the wrong thing, but potentially doing the wrong thing – like booking a flight based on faulty or incorrectly obtained facts.
• Latency: Refers to the speed or delay between a user asking a question and receiving a response from an agent, with “thinking” or process logic in between, often involving the apply of tools. This adds extra time compared to standalone LLMs or chatbots.

The most significant facts in the report:

• According to respondents, the cost of implementing AI agents is no longer crucial. 32% of them cite quality as the main barrier to adoption and implementation.
• Quality in this context refers to accuracy, consistency and avoidance of hallucinations.
• Meanwhile, there’s an intriguing catch: the second most critical barrier varies by company size, with petite startups citing delays, and companies with more than 2,000 employees citing security and compliance.

Similar supporting evidence can be found in the article cited earlier Report on barriers to the implementation of artificial intelligence by Deloitte, and detailed evidence on the main blockers to entrepreneurship could be further analyzed Average article.

Ivan Palomares Carrascosa is a thought leader, writer, speaker and advisor in the fields of Artificial Intelligence, Machine Learning, Deep Learning and LLM. Trains and advises others on the apply of artificial intelligence in the real world.