AI is revolutionizing software development, but intelligent agents present structural challenges
New research from the IBM Institute for Business Value reveals that developers spend 80% of their time understanding existing code rather than writing new code. This phenomenon, linked to the complexity of legacy systems, is driving the adoption of agentic coding, an approach that integrates artificial intelligence to accelerate development workflows.
Quick Answer
Agentic coding reduces the time spent understanding legacy code, accelerating modernization processes. AI agents, however, require specific infrastructures to manage memory and tools, and have predictable failure modes such as infinite loops and hallucinations in planning.
AI in sports analysis: the UFC case
Alon Cohen, CTO of UFC, illustrated how AI is transforming fight analysis, identifying decisive moments with millimetric precision. This practical application demonstrates how artificial intelligence can extract insights from complex data streams, offering competitive advantages in sectors not traditionally associated with technology.
The evolution of RAG: from simple retrieval to adaptive systems
Sam Anthony, senior researcher at IBM, outlined the evolution of Retrieval-Augmented Generation (RAG), moving from simple retrieval systems to frameworks capable of supporting complex decision-making processes. Semantic search and hybrid retrieval are becoming fundamental for managing multi-step searches, where AI must integrate information from different sources.
Agent OS: the infrastructure needed for reliable AI
Bri Kopecki highlighted how current AI agents operate without consistent memory, performing complex tasks such as booking flights or writing code without maintaining temporal context. The introduction of an operating system for agents (Agent OS) could solve these challenges, providing memory, tool management, and security guardrails.
Predictable failure modes of AI agents
Meenakshi Kodati classified the main causes of AI agent failures, including infinite loops in planning, hallucinations during task execution, and unsafe use of external tools. According to the researcher, these problems can be mitigated through more robust design and continuous performance monitoring.
The challenge of AI-generated mathematics
OpenAI and Meta are producing mathematical discoveries at an unprecedented rate, raising questions about how humans can verify and interpret these results. This scenario could lead to a fragmentation of knowledge, with AI generating more theory than can be validated.
The role of the Chief AI Officer in modern organizations
New data indicates that the role of the Chief AI Officer (CAIO) is becoming increasingly relevant, even if its mandate is still evolving. This key role requires specific skills in managing artificial intelligence projects, from design to operational implementation.
AI in cancer diagnosis
Researchers are training artificial intelligence models to detect immune activity within tumors using only standard histological slides. This approach could revolutionize early cancer diagnosis, offering more precise and accessible tools to doctors.
DeepSeek and the stability of LLM
A new approach developed by DeepSeek aims to improve the stability of training Large Language Models (LLM) at scale. This technological progress could facilitate the adoption of more powerful and reliable language models in business contexts.
The economic impact of agentic coding on companies
According to the latest estimates from McKinsey, the adoption of agentic coding could reduce the costs of maintaining legacy code by up to 40%, with a return on investment achieved in less than 18 months. This trend is pushing companies to re-evaluate their IT budgets, with a projected 20% increase in spending on SOC as a Service and MDR services to manage the security of new human-machine hybrid systems.
The legal challenges of AI-generated mathematics
The wave of mathematical discoveries produced by OpenAI and Meta raises crucial questions about copyright and intellectual property. In Europe, the Court of Justice of the EU is considering whether AI-generated results can be patented, while in the United States, the USPTO has already issued temporary guidelines requiring explicit declaration of the use of AI systems in patent applications.
The complexity of compliance for Chief AI Officers
The emergence of the Chief AI Officer role is creating new regulatory challenges, particularly regarding NIS2 compliance and the DORA regulation. According to a Deloitte report, 65% of companies with a CAIO are modifying their governance models to comply with new regulations, with particular attention to operational risk management and the protection of sensitive data.
Integrating AI into medical diagnostics: opportunities and risks
The use of AI models to analyze histological slides represents an unprecedented opportunity for personalized medicine, but also raises concerns about Data Loss Prevention and protection of sensitive data. Hospitals adopting these technologies will need to invest in advanced GDPR compliance and specialized cyber insurance to cover risks related to the processing of critical biomedical data.
The future of agentic coding: towards self-managed systems
IBM research indicates that by 2028, 70% of AI agents will operate on self-managed platforms, capable of self-diagnosis and self-repair. This development will require an evolution of cloud infrastructures, with a particular focus on cloud security posture and the implementation of zero trust architectures to prevent unauthorized access.
The evolution of the tech job market
The widespread adoption of agentic coding is transforming the job market, with a growing demand for profiles specialized in incident response and breach remediation. According to LinkedIn, job offers for roles related to AI agent management have increased by 150% in the last two years, with average salaries exceeding those of traditional software developers.
The challenge of technological standardization
One of the most significant obstacles to the large-scale adoption of AI agents is the lack of shared technological standards. IEEE is working on new guidelines for interoperability between different agentic coding platforms, while NIST has launched a project to define reliability and security metrics for these systems.
Implications for SMEs
Small and medium-sized enterprises are facing particular challenges in adopting agentic coding, with initial costs that can exceed €50,000 for the implementation of adequate infrastructures. However, as-a-service solutions such as managed Security Operations Center and scalable SIEM are making these technologies more accessible, lowering the minimum investment threshold.
The critical role of continuous training
The rapid evolution of agentic coding requires a significant commitment to the continuous training of technical staff. According to Gartner, 40% of companies adopting AI agents invest more than 10% of their training budget in specific courses for these technologies, with a particular focus on penetration testing and vulnerability assessment.
Future perspectives: towards a new era of software
The integration of AI into software development represents only the beginning of a broader transformation that will involve all aspects of technology. By 2030, it is predicted that 90% of corporate software will be developed or maintained with the support of intelligent agents, creating an ecosystem in which humans and machines will collaborate closely to solve complex problems.
Frequently Asked Questions
What is the average cost of implementing an agentic coding system?
The average cost of implementation ranges from €30,000 to €100,000, depending on the complexity of the system and existing infrastructures. As-a-service solutions can significantly reduce these initial costs.
What are the main risks associated with the use of AI agents?
The main risks include infinite loops, hallucinations in planning, and unsafe use of external tools. These problems can be mitigated through robust design and continuous monitoring.
How can I prepare my company for the adoption of agentic coding?
Start by evaluating your existing infrastructures and identifying areas where AI agents can bring the greatest benefits. Invest in continuous training for your staff and consider as-a-service solutions to reduce initial costs.
What is the role of a Chief AI Officer?
The Chief AI Officer is responsible for the strategy and implementation of artificial intelligence projects within an organization. This role requires specific skills in managing complex projects, from design to operational implementation.
Editorial Note and Disclaimer
The guides and content published on GoYou are the result of independent research and analysis activities, for informational, educational, and in-depth purposes.
GoYou does not constitute a journalistic publication or an editorial product pursuant to Law No. 62/2001 and does not provide real-time information.
The GoYou project does not provide professional, technical, legal, or financial advice and disclaims all responsibility for the improper use of the information published.
In the Crypto sector, every investment involves risks: readers are invited to always inform themselves independently before making any decision.