What Will AI-Driven Software Engineering Look Like in 2028?

In 2024 the author drafted a chapter on the future of "AI+Software Engineering" but recent breakthroughs—especially the FARS system that generated 118 papers in 273 continuous hours at a cost of about $1,034 per paper—show that the original predictions were overly conservative.

The Citrini Research "2028 Global Intelligence Crisis" scenario warns that, if current trends continue, AI will replace many white‑collar jobs, creating a "ghost GDP" where productivity soars while wages and consumption fall.

Key technical developments since the last forecast include the maturation of multi‑agent frameworks such as Microsoft AutoGen, the rise of AI‑Spec‑Driven Development (AI‑SDD), the emergence of new "Skills", and the rapid improvement of models like Claude 4.6, Gemini 3.1 Pro and Kimi 2.5. Tools such as Claude Code, OpenAI CodeX, Cursor, Qorder and Trae have accelerated the curve from linear to exponential growth.

2028 scenario : By the end of 2025, agentic programming tools will have achieved a step‑wise capability jump. In early 2026 a capable developer could recreate the core functionality of a medium‑scale SaaS product within weeks, fundamentally disrupting existing business models.

AI Intent‑Driven Development (AI‑IDD) will replace traditional AI‑SDD. Instead of merely translating specifications into code, AI will infer human intent from vague natural‑language requests, user behavior data, and multimodal inputs (sketches, voice, gestures). For example, an AI could be asked to build a TikTok‑like knowledge‑sharing platform and automatically generate architecture, code, APIs, and initial UI/UX, cutting prototype cycles from months to days.

Natural‑language intent : AI refines vague goals into detailed modules and interfaces.

User behavior insight : AI predicts latent needs and proposes new features.

Cross‑modal cues : Hand‑drawn sketches or eye‑gaze inform design decisions.

Multi‑agent "internal research society" will emerge, where specialized agents (Architecture Design Agent, Code Generation Agent, Test Agent, Optimisation Agent) negotiate, debate, and reach consensus on solutions. Test agents will not only run tests but also formulate hypotheses, design experiments, and produce reflective diagnostic reports, mirroring the FARS approach.

Autonomous testing and self‑healing will shift from defect detection to proactive prevention. AI security agents will continuously scan codebases, predict vulnerable versions, and auto‑patch them; performance agents will anticipate bottlenecks and re‑allocate resources before users notice latency; UI agents will verify cross‑device consistency via visual recognition.

Pre‑emptive vulnerability patching : Automatic detection and remediation of known library flaws.

Performance self‑healing : Predictive scaling and micro‑service refactoring.

Behavioral consistency checks : Visual UI verification across platforms.

Human‑AI symbiosis and role redefinition will create new positions such as "Master Agent" and "Intent Engineer" who define high‑level visions, guide AI collectives, and arbitrate conflicts among agents. "Soul Engineers" will embed corporate values, aesthetics and empathy into AI‑generated software, while "Ethics Architects" will enforce fairness, transparency and privacy, acting as the last safeguard against the predicted intelligence crisis.

Intent Engineer : Translates vague business goals into executable macro‑intent for AI.

Mentor/Arbiter : Guides AI agents, sets constraints, resolves ideological clashes.

Soul Engineer : Infuses non‑functional values (brand spirit, user emotion) into AI output.

Ethics Architect : Ensures compliance with fairness, privacy and societal responsibility.

Economic and social impact : The "ghost software factory" effect will erode traditional SaaS revenue streams as internal AI teams can replicate or surpass external SaaS offerings at a fraction of the cost. Development organisations will flatten, with a few high‑level agents directing massive AI workforces, and AI‑generated code and designs may become tradable "intellectual assets" subject to valuation and securitisation.

Legal and ethical challenges will arise around code ownership, liability for AI‑induced bugs, and potential discriminatory decisions made by optimisation agents.

Technological accelerators include the rise of "World Models" that enable causal reasoning and situational awareness, early manifestations of General AI capable of cross‑domain learning, and the evolution of programming paradigms toward intent‑oriented languages and highly intelligent low‑code/no‑code platforms.

In conclusion, by 2028 AI‑augmented software engineering will no longer be merely about efficiency gains; it will represent a profound restructuring of the software value chain, the redefinition of human roles, and a societal challenge that demands urgent attention to talent transformation, education reform and social safety nets.