- Home
- Insights
Notes from the work.Insights on technology due diligence, AI engineering and technology leadership
Written by the people who do the work: what we look for in a deal, how we build with AI, and what we've learned along the way.
All articles
OpenClaw: the open-source lobster and the UAL
What OpenClaw is, how it separates the messaging interface from the agent runtime, and how MCP and unified action layers fit together.
Umer QaisarSenior Fullstack EngineerThe Agent's Memory is a Product Decision
AI coding agents don’t always get things wrong because the model isn’t good enough. Sometimes, they simply have too much information to work with or the wrong information at the wrong time. In our experience, the difference comes down to how the agent’s memory is structured. Keep the basics that apply to every task in permanent memory, and move everything more specific into skills that can be brought in when needed. It sounds simple, but it can make a real difference: fewer surprises, more consistent work, and a smoother path from a client request to something that actually ships.
Sabeel AtherProduct and Revenue LeadThe 6 Types of Memory That Turn AI Agents Into Colleagues
AI agents need more than intelligence. They need memory. From understanding the current conversation to recalling past experiences, following company processes and remembering what needs to happen next, we explore the six types of memory that turn AI assistants into capable digital colleagues.
Hashim MazharPartner, Technology and ProductMoving Up the Knowledge Pyramid with AI
As AI takes over the Knowledge layer of the DIKW Pyramid, technology is shifting from simple data collection to synthesizing deep context. While traditional models connected the dots, they lacked the reasoning, context, and narrative required for true insight. By automating knowledge creation, AI frees investment and finance professionals from heavy data processing tasks. This enables human operators to focus on the Wisdom layer, applying high-level judgment and strategic decision-making.
Hashim MazharPartner, Technology and ProductSkills: the next evolution of prompting
As Artificial Intelligence evolved from simple prompts to complex context engineering, a critical new primitive has emerged for product teams: Agent Skills. Instead of rewriting prompts or stuffing system messages, teams can now encode their stabilized frameworks and judgment directly into reusable, self-activating folders. By shifting the focus from individual prompting to scalable organizational knowledge, this new approach allows a product team's collective expertise to compound over time.
Sabeel AtherProduct and Revenue LeadFrom Single-Model AI to Multi-Agent Systems: What Changed When We Adopted OpenAgent
What changed when our team moved from one model, one prompt, one task at a time to a multi-agent setup with Oh My OpenAgent, including the numbers and the trade-offs.
Aman TiwariProduct EngineerThe Unified Action Layer: How AI Agents Will Reshape Computing Integration
Computing's fundamental pattern is making complex capabilities consumable through abstraction layers. Today's fragmented AI integrations are poised for the next major consolidation through unified action layers, transforming how agents interact with digital infrastructure.
Hashim MazharPartner, Technology and ProductThe fundamentals haven't changed. AI just lets us fail faster.
Most takes on AI and product management get it backwards. They treat AI as a replacement for the work, when it's really a compression of the feedback cycle around the work. The skills that made a good PM in 2005 still make a good PM now. Customer empathy, prioritization, judgment, knowing your triangle. What's different is how fast you find out you're wrong. Three years ago, a bad assumption cost a sprint. Now it costs an afternoon. The job hasn't changed. The loop has. And that shift is quietly making the unglamorous parts of product management more valuable, not less.
Sabeel AtherProduct and Revenue LeadFrom Exam Problems to Production AI
Two months ago I was solving competitive exam problems. Today I'm shipping features on a production AI platform spanning React, FastAPI, dsPy, Temporal, and multi-LLM orchestration. This is an honest read on what those two months actually felt like, the first bug I broke with a TanStack Query optimistic update, the pgvector mismatch that taught me where real problems in AI systems hide, and the "obvious" question I asked in week two that unpacked a whole system design decision. The gap between knowing AI and shipping AI turned out to be smaller than I expected, but only because of the environment.
Aman TiwariProduct EngineerKnowledge Bases: A Leadership Guide
Knowledge bases are the backbone of every successful AI initiative, yet the term means different things to different people. This guide breaks down what knowledge bases really are, how they're built at every scale, and why getting them right is the single most important step before deploying LLMs, RAG, or agentic AI in your organisation.
Hashim MazharPartner, Technology and ProductHow to Use AI to Create a Winning Product Strategy
Strategy actually has two jobs: Creation (finding direction) and Management (validating and adjusting that direction). You need a clear workflow and feedback process, a "strategy loop", into which AI can fit, rather than trying to fit your strategy into AI. This ensures that you are compressing the necessary research without compressing the critical human thinking that provides the strategic foundation.
Sabeel AtherProduct and Revenue LeadThe End of Free Data: Private Intelligence and the Enterprises' Race to Lose
The era of building AI on publicly available web data is rapidly approaching its limits. As this "free" resource becomes exhausted, enterprises that can unlock and leverage their private data assets will gain an unprecedented strategic advantage over competitors still relying on generic models.
Hashim MazharPartner, Technology and ProductWritten by practitioners.
No ghostwriters and no content farm. Each article is written by someone who does this work.
Let's talk.
Read something that sounds like your situation? Tell us about it. The first call is free.