The Future of Software Engineering: Key Takeaways from an Agentic Programming Retreat

Introduction

Last week, a gathering of software development professionals convened to discuss the profession's trajectory in the era of agentic programming. The event operated under the Chatham House Rule, meaning insights can be shared but not attributed. Below are the most thought-provoking ideas that emerged, covering everything from rapid code porting to legacy modernization strategies.

The Future of Software Engineering: Key Takeaways from an Agentic Programming Retreat — Source: martinfowler.com

Rapid Code Porting with LLMs: The GNU Cobol Example

Behavioral Cloning in Three Days

One project showcased the power of large language models (LLMs) by creating a behavioral clone of the GNU Cobol compiler in Rust. The resulting codebase comprised 70,000 lines of Rust and was built in just three days. This demonstrates how LLMs can efficiently port existing code to new platforms, significantly reducing development time.

The Importance of Regression Tests

Such rapid porting relies heavily on robust regression tests to ensure behavioral fidelity. The speaker noted that while GNU Cobol's test suite quality was unknown, having a comprehensive test suite—or building one if an implementation exists—is crucial for validating the port.

Interrogatory LLMs: A New Approach to Spec Review

Large specification documents can be overwhelming for human reviewers. One attendee proposed an innovative solution: using an LLM to interview a human expert about the spec. This interrogatory LLM asks targeted questions to verify correctness, effectively turning the review into a dialogue that surfaces ambiguities and errors.

Understanding Organizational History through Change Control

A consultant shared a practical tip: the first thing they do when engaging with an organization is to read their change-control board guidelines. These guidelines represent the scar tissue of past failures and lessons learned. Understanding this history is key to grasping why systems are designed the way they are, a principle that applies broadly beyond AI.

Rethinking Legacy Modernization: The Case for Lift and Shift

The Old View: Avoid Feature Parity

Many modernizers have historically dismissed "lift and shift"—porting a legacy system to a new platform while preserving all features—as a missed opportunity. Legacy systems often accumulate unused features (reportedly 50% per a 2014 Standish Group study) and outdated processes. The old approach favored understanding current user needs and prioritizing them over preserving bloated functionality.

The New View: LLMs Change the Cost Equation

However, LLMs have drastically reduced the cost of migrating code. One attendee argued that lift and shift should now be the first step in any legacy modernization. The expense is no longer prohibitive, and moving to a modern platform makes subsequent evolution cheaper and safer. The caution: don't stop there; use the new environment to iteratively improve.

Financial Industry Challenges: Legacy, Regulation, and Risk

Several participants from the financial sector highlighted the unique pressures of operating complex legacy systems under strict regulatory controls. The risk of software errors with monetary consequences is high, making any migration approach need to balance innovation with safety. The insights above—especially around lift and shift—apply with extra caution in this domain.

Conclusion

The retreat underscored that agentic programming is reshaping core practices in software engineering. From efficient code porting to smarter spec reviews and a new perspective on legacy modernization, these ideas offer a glimpse into a future where LLMs augment human expertise. As one attendee wryly noted, even the most deeply entrenched practices can be re-evaluated when costs shift dramatically.

Tags: