Latest Trends in AI Coding

Last time I talked to The Australian Financial Review, it was just days after the ChatGPT meteor in November 2022. We had witnessed the global big bang of generative AI and GitHub Copilot was ready for take-off. Today, the majority of software developers are creating code with the help of AI tools. And I believe that’s a good thing. Three predictions of where we are going next: 1. 90% of code will be written by agents. Three years ago, Copilot was already creating 40% of code in those files where it was enabled, and I predicted that 80% of code would be written by AI in 2027. Today, this number is already exceeded by some developers that know how to get the most out of AI code generation. As such, I believe that number is about to rise even higher, with the mass deployment of agents at scale. 2. Agents will be the application layer of AGI. It’s only logical that we will only achieve AGI with a deep technology stack. In that stack, agents are the application layer, the interface through which human developers interact with models of greater and greater intelligence. With this in mind, AGI will arrive in software development tools before anywhere else. 3. We will need human developers. The implementation of agents doesn’t mean human developers will go away. Far from it. Your skill as a developer is no longer solely about developing code. In fact, it never has been. Software development is a craft, it’s about skill, care, precision, expertise, and pride. We will continue to build software that matters and they will use AI as one of the tools that complements our creativity. I’ve said it before, and I will keep saying it: There has never been a more exciting time to be a developer. And I don’t think I’m the “Godfather” of anything, but we’re certainly making an offering you can’t refuse. 😎 Thanks for the story, The Australian Financial Review!

90% of code written by developers using Windsurf’s agentic IDE is now generated by AI. This isn't science fiction. It's happening today. In 2022, auto-complete was revolutionary at 20-30% of code. Now we've entered the age of AI agents in software development. 7 ways agentic development environments are transforming coding today - with glimpses of tomorrow: 1️⃣ Unified Timeline (Now): Today's AI agents operate on a shared timeline with you, understanding your actions implicitly - viewing files, navigating code, and making edits without conflicting with your changes. 2️⃣ No More Copy-Paste (Now): Modern agent-based IDEs eliminate copy-pasting from chat windows. The agent lives where you work, seeing your context without you needing to explain it repeatedly. 3️⃣ Terminal Integration (Now): Commands run directly in your existing environment. When the agent installs a package, it goes to the same environment you're using - no more separate sandboxes. 4️⃣ Auto-Generated Memories (Now & Evolving): Leading AI development tools build memory banks of your preferences. Tell it once about your project architecture, and it remembers. By 2025, experts predict 99% of rules files will be unnecessary. 5️⃣ Implicit Documentation (Now & Evolving): Modern agents automatically detect your packages and dependencies, then find the right documentation without you needing to specify versions. 6️⃣ Beyond Context Prompting (Now & Evolving): The old '@file' and '@web' patterns are becoming obsolete. Today's advanced agents dynamically infer relationships between code and documents most of the time. 7️⃣ Future Vision (Coming Soon): Soon, agents will anticipate 10-30 steps ahead, writing unit tests before you finish functions and performing codebase-wide refactors from a single variable edit. The most striking realization: this isn't the future. It's happening now. When developers have agents that understand their implicit actions, remember their preferences, and improve with advancing models, productivity explodes. --- Are you still copy-pasting from ChatGPT, or have you embraced agentic development tools in your workflow? [Insights inspired by Kevin Hou's presentation at the AI Engineering Summit] --- Enjoyed this? 2 quick things: - Follow me for more AI automation insights - Share this a with teammate 

📋 I Treated My AI Coding Tools Like Interns—It Changed Everything When I shifted from "co-coding" with AI to managing AI tools as if they were junior developers, things clicked. I wrote specs. I chunked tasks. I made documentation a priority. Suddenly, AI-generated code was better aligned, easier to debug, and didn’t go rogue. It was like having a virtual dev team, but only if I acted like their Tech Lead. Now, the tooling to scale that model is arriving. Companies like Auctor, Cloobot, and Ressl AI are tackling requirements and architecture. Cirra AI automates Salesforce changes. TestZeus and Testsigma eliminate QA bottlenecks. SRE.ai (YC F24), Copado, Opsera, Hubbl Technologies, and Elements.cloud are handling deployments and DevOps at scale—with LLMs under the hood. 💡 The next wave of SDLC tooling is agentic. Are you ready for the agent era of software development? 👉🔗https://lnkd.in/g8fAtCDs #AIEngineering #CognitiveDevOps #AIAgents #SDLC #SoftwareDevelopment #LLMTools #AgentSwarm #DevOpsAutomation #SalesforceDevOps #GenAI #TechLeadership #AIinSoftware #VirtualDevelopers #AITooling #AIProjectManagement #AgentEconomy #FutureOfDevOps #AIProductivity 

🚀 AI Agents: 4 Trends to Watch in 2025🌍💡 AI agents are revolutionizing industries, moving beyond copilots to autonomous digital workers 🤖. As we enter 2025, four key trends are shaping the AI agent landscape: 1️⃣ Big Tech & LLM Developers Dominate General-Purpose Agents 🔹 Tech giants (OpenAI, Anthropic, etc.) are driving AI advancements, making agents cheaper, more powerful, and widely available. 🔹 400M weekly users on ChatGPT showcase the massive distribution advantage. 🔹 Enterprise adoption is increasing, but big tech’s dominance pressures startups to specialize. 2️⃣ Private AI Agent Market Moves Toward Specialization 🔹 Horizontal AI applications (customer support, software development) are crowded – differentiation is key. 🔹 Industry-specific AI agents in healthcare, finance, compliance, and logistics are poised for growth. 🔹 Deeper workflow integrations & leveraging proprietary data will create competitive moats. 3️⃣ AI Agent Infrastructure Stack Crystallizes 🔹 The AI agent ecosystem is evolving into a structured stack with specialized solutions: ✅ Data curation (LlamaIndex, Unstructured) ✅ Web search & tool use (Browserbase) ✅ Evaluation & observability (Langfuse, Coval) ✅ Full-stack AI agent development platforms gaining traction 4️⃣ Enterprises Shift from Experimentation to Implementation 🔹 63% of enterprises place high importance on AI agents in 2025. 🔹 Challenges remain: Reliability & security (47%), Implementation (41%), Talent gaps (35%). 🔹 Solutions: Human-in-the-loop oversight, stronger data infrastructure, and enterprise-grade agent platforms. 🚀 2025 is a breakout year for AI agents – the shift from copilots to autonomous digital workers is happening now! 📈 #AIAgents

If you were hoping for a slowdown in AI innovation in 2025, the first 38 days of the year are proving that the space is only accelerating. My six predictions for AI and software engineering this year - backed by what we're seeing in the market today: 1. The LLM moat is shrinking - With DeepSeek approaching closed models and available for free, value is shifting to what you build on top. Basic LLM access is becoming more of a commodity - and that's good for innovation. 2. Enterprise AI will go vertical - The next wave isn't general-purpose models. It's specialized AIs trained on proprietary enterprise data. Every major industry will build domain-specific models on open source foundations. 3. Software engineering teams will grow, not shrink - Controversial take: AI making software development cheaper and more predictable will increase demand for engineers. Smart CTOs are using AI to tackle their feature backlog, not reduce headcount. 4. RAG trumps fine-tuning - Real-time context beats static training. The future is retrieval-first: lower costs, better security, instant updates. 5. Two AI-assisted programming paradigms evolve - Engineers will seamlessly switch between: Direct coding with AI assistance and Meta-programming through natural language. The key is having tools that maintain context across both modes. 6. AI agents for software get real - Beyond code completion and chat, AI will handle: Test generation, migrations, security scanning, documentation, more complex refactors. But with human oversight, not autonomously. Augment Code https://lnkd.in/eerVneuX

People often ask me how AI is changing the space that Statsig operates in. For most of the last 3 years, my answer has been "not much". Sure, most of the companies that build leading AI products rely on Statsig - but for the most part, AI hasn't dramatically altered our products or the way they're used. Today, that's starting to change. The biggest trend I've seen is that AI has gone from something that was only applied in a few leading companies to something used by everyone. Case in point: last month, when we announced offline prompt evaluations and reached out for early design partners, the response wasn't just from AI-native companies. We heard from leading retailers, fintechs/neobanks, health-tech companies, and enterprise software providers. Of course, these companies aren't just building AI into their product - they're using AI to build their product. Cursor, Claude Code, Codex and Devin are now a core part of most companies' development stack. So how is this changing our space? Siddharth and I pulled some thoughts together around 4 core themes... 1. Traditional offline testing is being replaced by evals: Rather that leveraging traditional offline testing workflows (i.e., looking at model outputs vs. benchmarking datasets), product teams run tests vs. their own small eval datasets. Then, once they're confident nothing is broken, they ship changes to production as an A/B test to get real user feedback. 2. More AI-generated code = more need for experimentation: Anthropic’s CPO recently shared on Lenny’s that 90% of their code is now AI-generated. As more code gets generated without direct oversight, there's a greater need for quantitative measurement of what's working (and what isn't). 3. Every engineer is becoming a growth engineer: With AI, you can now test 10x more growth hacks. Building is becoming easier and easier - and figuring out which ideas could actually move your product metrics is becoming even more important. 4. Unique, context-rich data is the differentiator for great AI apps: As LLMs get better (and more commoditized) the thing that differentiates one app from another is thoughtful product decisions + unique data. This means adopting AI, but in a way that is very sensitive to driving great user outcomes. And this starts with understanding what a "great user outcome" means - and how you can measure it. Super interested to see how this changes over time - particularly as the eval stack changes with more and more agents. Full blog below ⬇️ https://lnkd.in/gk3dw_FY

The Future of AI Belongs to the Prepared. If you want to stay relevant in 2025 and beyond, mastering foundational AI skills is no longer optional. That’s why I created this visual: “15 AI Skills to Master in 2025”—a roadmap for developers, data engineers, and tech leaders navigating the GenAI era. Here’s what the future demands: ⫸ Prompt Engineering – Still the secret sauce to great LLM output. ⫸ AI Workflow Automation – No-code and low-code tools will drive faster innovation. ⫸ AI Agents & Agent Frameworks – LangChain, CrewAI, AutoGen… Agentic AI is the new operating model. ⫸ RAG (Retrieval-Augmented Generation) – Combine LLMs with private data sources for real-time intelligence. ⫸ Multimodal AI – Text, code, images, audio… future models speak every language. ⫸ Custom LLMs & Fine-Tuning – Build assistants fine-tuned for your use case. ⫸ LLM Evaluation & Observability – If you can’t measure it, you can’t improve it. ⫸ AI Tool Stacking – Combine APIs and agents into powerful workflows. ⫸ SaaS AI App Development – AI-native products require scalable infra and modular thinking. ⫸ Model Context Protocols (MCP) – Handle memory, context, and token budgeting across agents. ⫸ Autonomous Planning & Reasoning – ReAct, ToT, and Plan-and-Execute are no longer just research. ⫸ API Integration with LLMs – Connect the real world to your AI agents. ⫸ Custom Embeddings & Vector Search – Semantic search is foundational to personalization. ⫸ AI Governance & Safety – As AI grows, so do the risks. Guardrails are critical. ⫸ Staying Ahead with AI Trends – Read, build, share, repeat. Constant learning is non-negotiable. Whether you’re building the next intelligent platform or leveling up your career, this roadmap outlines what matters most. Use it to audit your current skillset. :-)

Folks interested in AI / AI PM, I recommend watching this recent session by the awesome Aishwarya Naresh Reganti talking about Gen AI Trends. ANR is a "Top Voice" that I follow regularly, leverage her awesome GitHub repository, consume her Instagram shorts like candy and looking forward to her upcoming Maven Course on AI Engineering. https://lnkd.in/g4DiZXBU Aishwarya highlights the growing importance of prompt engineering, particularly goal engineering, where AI agents break down complex tasks into smaller steps and self-prompt to achieve higher-order goals. This trend reduces the need for users to have extensive prompt engineering skills. In the model layer, she discusses the rise of small language models (SLMs) that achieve impressive performance with less computational power, often through knowledge distillation from larger models. Multimodal foundation models are also gaining traction, with research focusing on integrating text, images, videos, and audio seamlessly. Aishwarya emphasizes Retrieval Augmented Generation (RAG) as a successful application of LLMs in the enterprise. She notes ongoing research to improve RAG's efficiency and accuracy, including better retrieval methods and noise handling. AI agents are discussed in detail, with a focus on their potential and current limitations in real-world deployments. Finally, Aishwarya provides advice for staying updated on AI research, recommending focusing on reliable sources like Hugging Face and prioritizing papers relevant to one's specific interests. She also touches upon the evolving concept of "trust scores" for AI models and the importance of actionable evaluation metrics. Key Takeaways: Goal Engineering: AI agents are learning to break down complex tasks into smaller steps, reducing the need for users to have extensive prompt engineering skills. Small Language Models (SLMs): SLMs are achieving impressive performance with less computational power, often by learning from larger models. Multimodal Foundation Models: These models are integrating text, images, videos, and audio seamlessly. Retrieval Augmented Generation (RAG): RAG is a key application of LLMs in the enterprise, with ongoing research to improve its efficiency and accuracy. AI Agents: AI agents have great potential but face limitations in real-world deployments due to challenges like novelty and evolution. Staying Updated: Focus on reliable sources like Hugging Face and prioritize papers relevant to your interests. 🤔 Trust Scores: The concept of "trust scores" for AI models is evolving, emphasizing the importance of actionable evaluation metrics. 📏 Context Length: Models can now handle much larger amounts of input text, enabling more complex tasks. 💰 Cost: The cost of using AI models is decreasing, making fine-tuning more accessible. 📚 Modularity: The trend is moving towards using multiple smaller AI models working together instead of one large model.

25% of Y Combinator’s latest startups have codebases written entirely by AI - not AI-assisted, but AI-authored. Add to that: Statista’s latest data reveals how AI is deeply embedded across dev workflows worldwide: ➡ 82% of devs use AI for code generation ➡ 56.7% for debugging and help ➡ 40.1%+ for documenting code And the broader industry paints the same picture: ➡ Google reports that over 30% of its new code is now AI-generated ➡ Developers using AI tools like GitHub Copilot report up to 55% productivity gains ➡ 90% of U.S. developers and 81% of Indian developers perceived an increase in code quality when using AI coding tools What does this mean for the industry? Development workflows are evolving into conversational, prompt-driven processes, and SDKs, APIs, and documentation must evolve for an AI-first world. ✅ The reality is clear: The future of software development is AI-driven, prompt-powered, and vastly faster. If you’re building developer tools, managing teams, or owning integrations, it’s time to rethink your strategy. Ask yourself: ✅ How is AI changing your dev workflows? ✅ Are your teams prepared to leverage AI to shift focus from coding to strategy? ✅ Is your product ready to be consumed, understood, and extended by AI? #AI #Dev #AIDevelopment #AIBuilders

Wouldn’t it be great if we had a crystal ball to help us foresee the tech trends most likely to impact us in the immediate future? While we may lack the powers of divination, we do have the next best thing: our new 2024 Technology Trends Outlook! Here are some of my takeaways: - Job postings related to generative AI increased by 111%. Organizations are now focusing on scaling and expanding their internal capabilities, leading to a sharp increase in demand for data scientists, software engineers, and data engineers. - Generative AI investments surged 7x due to major advancements in text, image, and video generation. While gen AI adoption increased across various sectors, the technology, media, and telecommunications sector notably emerged as a leader. - New versions of AI-powered development tools are transitioning from proof of concept to wide-scale application. The software development industry has witnessed a significant turning point in the past year with the release of new versions of advanced AI-powered tools that are transforming the landscape. To learn more, check out the trends report here: https://lnkd.in/eBBjVF8e #McKinseyDigital #GenerativeAI #TechTrends2024