ICSE 2026 vs 2016

A Decade of Software Engineering Research — How Topics Have Shifted

321 ICSE 2026 Papers
101 ICSE 2016 Papers
3.2× Growth Factor
9 Topic Categories

Executive Summary

The AI Takeover

34.3%

of ICSE 2026 papers involve AI/ML topics, up from 6.9% in 2016. LLMs dominate with 75 papers on code generation and agents alone.

Security Surge

10.9%

of 2026 papers address security (up from 5.9%). Smart contract security emerged as an entirely new subcategory with 9 papers.

Testing Holds Steady

~14%

Testing remains a core pillar in both years, but fuzzing grew from 0 to 18 papers and domain-specific testing now covers autonomous driving and XR.

Classical SE Declines

↓50%

Program analysis, verification, and symbolic execution shrank from 18.8% to 8.7% of papers. Traditional SE topics are being absorbed by AI approaches.

Methodology

Data Collection

  • ICSE 2026: 321 paper titles extracted from the official ICSE 2026 Research Track page on researchr.org
  • ICSE 2016: 101 paper titles extracted via the DBLP API for the ICSE 2016 proceedings
  • Categorization: Each paper was manually assigned to one primary subcategory based on the dominant theme in its title. Categories were designed to capture both 2016-era and 2026-era topics.
  • Normalization: Because the conferences have different sizes (321 vs 101 papers), all comparisons use percentages rather than raw counts.

Limitations

  • Categorization is based on titles only, not abstracts or full papers, which may miss nuances
  • Some papers span multiple categories; each was assigned to one primary category using best judgment
  • ICSE 2016 had ~101 research track papers while ICSE 2026 accepted ~321, reflecting broader growth in the field
  • The 9 main categories with 33 subcategories were designed post-hoc to best capture both eras

Overall Research Landscape

The most striking change over the past decade is the explosion of AI & Machine Learning research, which grew from 6.9% to 34.3% of all papers. Meanwhile, Program Analysis & Verification and SE Process & People have seen their relative share decline, even as absolute paper counts grew. Testing remains a bedrock of the community, and Security has seen meaningful growth driven by smart contracts and supply chain concerns.

Main Category Distribution: 2016 vs 2026

ICSE 2026 ICSE 2016

Radar: Topic Profile Overlay

Percentage-Point Change (2016 → 2026)

Category Breakdown

Category 2016 (#) 2016 (%) 2026 (#) 2026 (%) Change

The AI Revolution in Software Engineering

In 2016, AI-related SE research consisted of 7 papers (6.9%) focused primarily on traditional ML techniques like defect prediction with word embeddings, statistical API learning, and early program synthesis with natural language. By 2026, this category has exploded to 110 papers (34.3%), driven almost entirely by the rise of Large Language Models.

AI & ML Subcategory Breakdown

Key Findings

LLM Code Generation & Repair is the single largest subcategory in all of ICSE 2026 with 28 papers (8.7%), covering code completion, program repair, code translation, and automated programming with LLMs.

LLM Evaluation & Analysis has emerged as its own research area with 27 papers (8.4%), investigating robustness, memorization, hallucination, fairness, and the limits of LLM capabilities.

LLM Agents & Autonomous SE represents a brand-new paradigm with 20 papers (6.2%), exploring multi-agent systems, SWE-bench benchmarks, and fully autonomous software engineering agents.

AI for Security accounts for 18 papers (5.6%), applying LLMs to vulnerability detection, jailbreaking, and AI supply chain security—a topic that simply didn't exist in 2016.

Topic Shifts: What Grew, Shrank, and Emerged

Topic Trajectory: 2016 → 2026

Growing Topics

AI & Machine Learning (+27.4 pp) saw the largest absolute growth, driven by the LLM revolution. Security (+5.0 pp) grew significantly with the emergence of smart contract security and supply chain attacks. Fuzzing went from essentially zero papers to 18 papers, becoming a dominant testing technique.

Stable Topics

Software Testing (~14%) maintained its share even as its composition shifted toward fuzzing and domain-specific testing. Software Maintenance & Evolution (~6-13%) remained present but with a shift toward dependency management over code review.

Declining Topics (Relative Share)

Program Analysis & Verification dropped from 18.8% to 8.7%—not because there are fewer papers, but because AI research has grown disproportionately. SE Process & People declined from 15.8% to 8.1%. Symbolic Execution went from 6 papers (5.9%) to just 2 papers (0.6%), a dramatic shift as LLM-based approaches subsume this space.

New-in-2026 Topics

Subcategory 2026 Papers Notes
LLM Code Generation & Repair 28 Entire subcategory new; LLMs didn't exist in 2016
LLM Agents & Autonomous SE 20 Multi-agent, SWE-bench, autonomous debugging
LLM Evaluation & Analysis 27 Robustness, memorization, hallucination research
AI for Security 18 LLM-based vuln detection, jailbreaking, AI supply chain
Smart Contract Security 9 Ethereum/DeFi security emerged post-2016
Supply Chain Security 8 SolarWinds, Log4j era concerns
Cloud & Microservices 3 Serverless, microservices architecture

Subcategory Deep Dives

Each main category contains multiple subcategories. The donut charts below show the internal composition of ICSE 2026 papers within each main category, revealing where the community's attention is concentrated.

AI & ML Subcategories (2026)

Testing Subcategories (2026)

Security Subcategories (2026)

Program Analysis Subcategories (2026)

Maintenance Subcategories (2026)

SE Process Subcategories (2026)

Requirements & Architecture (2026)

Specialized Domains (2026)

Paper Lists by Category

Click any category or subcategory below to expand and see the individual paper titles assigned to it.

Conclusion

The decade from ICSE 2016 to ICSE 2026 tells a dramatic story of transformation in software engineering research:

1. The LLM Revolution is Real and Pervasive. Over one-third of ICSE 2026 papers involve AI/ML, with LLMs appearing not just as research subjects but as tools across testing, security, program analysis, and maintenance. The emergence of "LLM Agents" as autonomous software engineers represents a paradigm shift that was unimaginable in 2016.

2. Security Has Moved Center Stage. From smart contract exploits to supply chain attacks (SolarWinds, Log4j, XZ Utils), security concerns now permeate SE research. The convergence of AI and security—both using AI for security and securing AI systems—is a defining theme of 2026.

3. Classical Techniques Are Being Augmented, Not Replaced. While symbolic execution and traditional static analysis have declined as standalone topics, they increasingly appear as components within LLM-assisted approaches. The field isn't abandoning formal methods—it's hybridizing them.

4. The Conference Has Grown Substantially. With 321 papers vs 101, ICSE 2026 is more than three times the size of 2016, reflecting both the field's growth and increasing submission volumes driven by AI-assisted research.

5. New Application Domains Have Emerged. Autonomous driving, extended reality, DeFi smart contracts, and RISC-V firmware—these domains barely existed in SE research a decade ago and now have dedicated testing and verification research streams.