AI

Agentic AI is fully mainstream in July 2026. The framework connecting everything: is enterprise ERP finally ready for agents that operate it autonomously?

AI features in mobile apps everywhere. But usage data tells a different story than what the hype suggests. Here's what people are actually using and paying for.

OpenAI o3, Claude's extended thinking, reasoning models are mainstream. They're not just harder problems solved faster—they enable entirely new categories of applications.

Year-end 2025. Not generic predictions about AI. Concrete, specific things I actually expect to see happening in enterprise software and mobile in 2026.

On-device AI models (Apple Intelligence, Google Gemini Nano) make offline-first AI possible. Real patterns for building features that work without internet.

Running LLMs in production is harder than demos suggest. Cost, latency, hallucinations, and liability issues that nobody warned you about beforehand.

AI needs clean data, but ERP data is usually a mess. Most companies can't fix data strategy fast enough to power real AI features. Here's what to do.

Single agents are useful, but multiple agents coordinating across complex workflows and making decisions together is where enterprise AI is truly headed.

The promise: fully autonomous ERP workflows with no human touch. The reality: partially autonomous, heavily guardrailed, where constraints matter more than AI.

Three years after ChatGPT launched, enterprise AI adoption is real but much narrower and messier than anyone predicted. What deployed, what didn't, why.

Chatbots are table stakes now. The real frontier is AI agents that autonomously take sequences of actions on behalf of users. Here's how to build it.

Every ERP vendor launched an 'AI copilot' in 2024. I've tested most of them. Here's what separates genuine workflow assistance from rebranded autocomplete.

2024 was the year AI agents shifted from research projects to true production reality. Here's what this fundamental shift means for the entire industry.

Phi-3, Gemma, Llama 3.1. In September, the market shifted from massive models to small, efficient ones. It's not a compromise—it's legitimately better.

Users now expect AI everywhere. Mobile apps without natural language interfaces feel outdated. The question isn't whether to add AI—it's how to do it.

RAG solves the hallucination problem that makes LLMs unreliable. But most implementations miss the architectural insight that makes RAG actually work.

Apple and Google are both shipping on-device AI. It's not about convenience. It's about control, privacy, and the end of cloud dependency for certain tasks.

Anthropic shipped Claude 3 in March with Opus, Sonnet, and Haiku tiers. For the first time, an LLM platform feels genuinely built for enterprise software.

A manufacturing client was hemorrhaging money from inventory mismanagement. We didn't just fix their ERP—we rebuilt it with AI understanding at the center.

AI isn't coming to ERP anymore—it's already here. I'm seeing enterprise systems that treat AI as a genuine first-class system citizen, not an afterthought.

Looking back at 2023: it's the year AI features landed in mainstream apps. Siri got better, Bard integrated with Google apps, ChatGPT app shipped. What it means for mobile.

Connecting LLMs to ERP workflows. Patterns for automating document processing, approvals, and data entry with AI inside your business system.

Meta's Llama 2 arrived in July. Open-source LLMs are suddenly competitive with commercial models. Here's what that means for businesses.

LLM hype is at maximum. Every startup claims AI. Here's a framework for figuring out which LLM applications are genuinely useful and which are just shiny.

Post-ChatGPT, everyone wants AI in their app. Here's what actually works on mobile, what doesn't, and what users actually want.

GPT-4 shipped in March 2023. The capability jump from GPT-3.5 is real. Here's what actually changed for those of us building products.

Every ERP client asking 'how do we add AI?' Here's where AI integration actually makes sense in your ERP, and where it's just expensive noise.

Six weeks after ChatGPT's launch, the frenzy is settling. Real capabilities are emerging from the noise, and they're remarkably different from what headlines promised.

ChatGPT launched November 30. I tested it relentlessly. Most discussion from people who haven't used it. Here's what I learned from real work.

Responsible AI became a buzzword. Usually compliance theater. EU AI Act forced real questions. What does responsible AI look like when building real products?

DALL-E 2 shipped in April 2022. But this moment isn't about art. It's about product design changing. Every product that produces something became a potential AI application.

Text-to-image AI exploded in 2022. DALL-E 2, Midjourney, Stable Diffusion shifted what's possible. Generative AI isn't about art—it's about automation at the content layer.

Facebook becoming Meta has everyone talking metaverse. Here's what's real, what's marketing, and why most of the discussion completely misses the actual opportunity.

We've been using Copilot since the technical preview dropped. Here's what actually happened—the productivity gains, the failures, and what's overblown.

DALL-E and CLIP changed what's possible with AI. Here's why combining vision and language is about to transform enterprise applications.

AI-assisted coding is no longer science fiction. GitHub Copilot is months away and developers need to start thinking about what it means now.

June hype around GPT-3 cooled. Real limitations emerged. What GPT-3 can actually do versus what people hoped became clear. Here's the honest assessment.

GPT-3 API access started in June. The demos going viral on Twitter made me reconsider what's achievable in AI-driven software within the next five years.

That impressive AI demo works great on curated data in a demo environment. Here's why it typically breaks when you deploy it into messy enterprise systems.

Amazon's biased recruiting AI made headlines last year. The problem is worse than most realize. Here's what companies need to know before deploying AI.

BERT and GPT demonstrated transfer learning's power. Building AI models no longer requires billions in compute. Here's what fundamentally changed in 2019.

OpenAI's staged GPT-2 release in Feb 2019 was genius marketing or genuine caution. Six months later, it's clear why this decision mattered significantly.

Most enterprise AI initiatives die quietly. I've watched dozens fail. Here's what truly separates the winners from the graveyard of abandoned ML projects.

OpenAI published GPT-1 in June. Most people missed it. I didn't. Here's why I think transformers are about to reshape everything we build.

Google's AutoML launched this year. Suddenly ML is accessible to non-experts. Is this genuine democratization, or just marketing genius? Here's my honest take.

ELMo launched in February. The NLP revolution is accelerating. Here's how language models will transform how enterprises find and process information.

TensorFlow 1.0 just launched. Meanwhile, most enterprises are still treating ML as a research curiosity. Here's the gap I see widening.