Better models alone will not solve agent reliability.
If an agent cannot remember, verify, recover, or be inspected properly, the problem is often the harness, not the model.
Wrote up my thinking on harness engineering:
www.anup.io/harness-engi...
If you're trying to standardise how your team uses AI across the full software development lifecycle, this repo template is worth a look.
github.com/pangon/ai-sd...
Claude Code as a self-managing AI team; one repo, multiple specialised agents (PM, architect, engineer, reviewer) coordinating autonomously. The agentic SDLC is here.
github.com/CronusL-1141...
It is the season of lists, so I looked at AI through Santa’s ledger instead of another hype story. This year was anything but simple.
www.anup.io/ais-naughty-...
#AIengineering #LLM #AIethics
Happy holidays, everyone!
The AI coding advice that works for someone else probably won't work for you. Not because they're wrong, but because the tools aren't stable enough for best practices yet.
www.anup.io/why-ai-codin...
#AIEngineering #LLMs #SoftwareDevelopment #AICoding
Scaling RAG in production isn't an LLM problem. It's a systems problem wrapped around a language model.
www.anup.io/rag-at-scale...
Imagine a UI generated on the fly for every query. That’s the idea behind Google’s Generative UI paper. I unpack the architecture, the prompt, and why this shift matters for agent-first applications.
www.anup.io/understandin...
Senior developers are better at using AI agents than junior developers.
Not because they code faster. Because they already know how to delegate, communicate clearly, and spot when something looks right but isn't.
People management = AI management.
www.anup.io/senior-devel...
Nicely done @shashanktomar.bsky.social
Reading time: 2–4 days.
Which, to be fair, is still faster than fine-tuning.
huggingface.co/spaces/Huggi...
Fine-tuning dropped from hottest trend to <10% of AI workloads in two years. Now it's coming back. Not as hype, but as the last 10% that actually matters.
Turns out some problems need you to teach the model your language, not just describe it in a prompt.
www.anup.io/fine-tuning-...
Stop trying to understand research papers in one read.
Pass 1: Abstract + intro + conclusion (is this even relevant?)
Pass 2: Full read, flag gaps
Pass 3: Revisit with fresh context
It's a workflow, not a puzzle. Treat it like one.
www.anup.io/how-to-read-...
Built a crisis support voice AI in 90 minutes at a hackathon and won. Full breakdown of the build + 7 tactics for shipping fast.
www.anup.io/eliza-redux-...
I love how Claude is so wholesome that it literally tells me to use a competitor’s product. 😂
Peak confidence. Peak alignment.
Want to know the difference between MCP and A2A checkout this talk delivered at the London AI Engineers meetup
www.anup.io/mcp-a2a-the-...
#MCP #A2A #AIProtocols
So true! Software release on Friday 😁😂🤣
www.youtube.com/shorts/Iojxt...
I tested ChatGPT Atlas vs Perplexity Comet on real automation workflows. Atlas won on reliability, so I'm switching over for now.
anup.io/atlas-unshru...
#atlas #openai #Perplexity #comet
