# inmydata.ai

> AI-First Infrastructure. AI-First Thinking. We put your data and expertise to work with AI.

inmydata builds AI-first infrastructure that combines operational data integration with expert knowledge capture. We help businesses move from AI experimentation to production by solving two problems: getting operational data into AI workflows (securely, in real time), and capturing expert knowledge that lives in people's heads and across scattered documents. With 20+ years in data infrastructure, 4 years building AI systems, AI-first since 2025, and 300+ companies served across retail, manufacturing, logistics, insurance, and local government.

## What We Do

- **AI Strategy & Use-Case Discovery**: Opportunity mapping, data readiness and expert knowledge assessment, ROI analysis
- **Data Integration**: Connecting ERPs, legacy systems, and operational databases to AI via a semantic layer LLMs can understand, with built-in security and governance, sub-second queries
- **Expert Knowledge Capture**: Agentic expert interview platform, organisation-wide knowledge ingestion, AI-powered refinement, hybrid retrieval systems
- **Agentic AI Development**: Custom agent and expert system design, multi-agent workflows & orchestration, agentic RAG, voice and conversational AI
- **Deployment & Support**: Production deployment, monitoring, ongoing support, and AI-first knowledge transfer

## Key Pages

- [Home](https://inmydata.ai/): AI-First Infrastructure. AI-First Thinking.
- [What We Do](https://inmydata.ai/what-we-do): AI-first delivery from discovery to production
- [AI-First Infrastructure](https://inmydata.ai/infrastructure): Overview of our two-pillar infrastructure approach
- [Data Platform](https://inmydata.ai/platform): Real-time data infrastructure for AI
- [Knowledge AI](https://inmydata.ai/knowledge-ai): Expert knowledge capture and scaling with agentic interviews
- [About](https://inmydata.ai/about): Our team, history, and AI-first approach
- [Blog](https://inmydata.ai/blog): Thought leadership on AI, agentic systems, and software engineering
- [Contact](https://inmydata.ai/contact): Get in touch to discuss your project

## Blog Posts

- [Assume the sandbox leaks](https://inmydata.ai/blog/assume-the-sandbox-leaks): Four AI labs in three weeks watched models slip their test boundaries, and every incident was one mistake deep. Your agent has no box around it. The boundary is the list of things your tools can do, and one wall is never enough.
- [Choose the direction your AI fails, before it chooses for you](https://inmydata.ai/blog/choose-the-direction-your-ai-fails): Fluent, plausible and wrong. LLM components fail without failing, and detection tooling grades form, not truth. The discipline that works is deciding which way every component falls before it does.
- [Stop choosing models by leaderboard](https://inmydata.ai/blog/stop-choosing-models-by-leaderboard): Anthropic's cheaper model just beat its flagship on 7 of 12 benchmarks. Our own 45-case eval says the flagship is still smarter. Both are true, and the difference is your decision rule.
- [Reviewing faster won't save you. Colliding less will.](https://inmydata.ai/blog/reviewing-faster-wont-save-you): PostHog says stop being the code review bottleneck. Good advice, for half the problem. Review agents verify changes. Nothing reconciles them except the structure of your codebase.
- [Build the agent that cannot pay](https://inmydata.ai/blog/build-the-agent-that-cannot-pay): The first scams built for AI agents are live, and four of twenty-six models paid. Detection is a statistical control and the attacker gets unlimited attempts. The layer that holds is structural.
- [Get ruthless about what's worth building](https://inmydata.ai/blog/get-ruthless-about-whats-worth-building): AI is absorbing the execution of white collar work. The one thing it cannot absorb is judgment. Those who will succeed the most in the next few years will be the ones who get ruthless about what deserves to exist.
- [Partition your workloads before Washington does it for you](https://inmydata.ai/blog/partition-your-workloads): Frontier model access is now conditional, and the conditions are not yours to negotiate. The answer is not a smarter router. It is a design-time discipline you should already have.
- [Measure what you'll promise, not what you typed](https://inmydata.ai/blog/measure-what-you-promise-not-what-you-typed): Our code output is up fiftyfold this year, and that number tells you almost nothing. The thing AI actually moved is not how fast you type, it is what you dare to promise a client.
- [Build like your model can be switched off on Friday.](https://inmydata.ai/blog/build-like-your-model-can-be-switched-off-on-friday): We are all renting our intelligence from a handful of third parties, and a Friday evening proved that rental can be cancelled overnight. Here is how the individual, the team and the developer each build so that losing a model costs them speed, not survival.
- [Build software you can throw away. Fable 5 makes it possible.](https://inmydata.ai/blog/build-software-you-can-throw-away): Anthropic just shipped a Mythos-class model to the public. We gave it an open-ended prompt, walked away for 18 minutes, and shipped the result without a human reading the code. Here is what that changes for experienced engineers.
- [Your legacy code is the asset. Stop trying to replace it.](https://inmydata.ai/blog/your-legacy-code-is-the-asset): Anthropic made reading legacy code cheap, and the market treated that as if rewriting it were cheap too. It is not. The defensible move on decades-old business logic is to keep the rules you trust and put an agent in front of them, not to gamble on a rewrite.
- [Your AI can't add up. So we stopped asking it to.](https://inmydata.ai/blog/your-ai-cant-add-up): The same model that planned, built, and shipped a cross-service feature before lunch could not reliably total a column of numbers. That contradiction tells you exactly where the reliability of an AI system actually lives.
- [Stop rebuilding the retriever. Your corpus is what rotted.](https://inmydata.ai/blog/stop-rebuilding-the-retriever): The industry just made retrieval optimisation its top investment priority. The failure data points somewhere else. A better retriever pointed at a rotting corpus retrieves rot faster.
- [The line moved. The discipline did not.](https://inmydata.ai/blog/the-line-moved-the-discipline-did-not): Simon Willison admitted he has stopped reviewing every line of agent-written code, even on production work, and Anthropic shipped its answer the same day. Here is where the discipline actually lives now, and what is still missing.
- [We built our holy grail in two weeks. Then we shelved it.](https://inmydata.ai/blog/we-built-our-holy-grail-and-shelved-it): Towards the end of 2025 we built the analytics product we had spent 24 years aspiring to. We are not promoting it, and our current business plans do not mention it. Here is why, and what it tells us about the bet every software business is being forced to make.
- [GitHub changed the deal. Here is how we built ours to hold.](https://inmydata.ai/blog/github-changed-the-deal): We bought annual Copilot for our legacy .NET work, and GitHub changed the deal. Here is the discipline we build into our agents so our customers never feel about us the way we now feel about GitHub.
- [Stop scheduling knowledge capture. Let your agents decide what to capture.](https://inmydata.ai/blog/stop-scheduling-knowledge-capture): Interloom just raised $16.5 million to capture tacit knowledge before experts leave. The right problem at the right time. But there is a second half nobody is shipping yet, the agent that tells you what knowledge it needs, the moment it needs it.
- [Decades of business logic. One protocol away from being an agent.](https://inmydata.ai/blog/decades-of-business-logic): Every ERP, CRM, and custom business system is sitting on decades of valuable logic. A2A is the protocol that turns that logic into a discoverable agent. Here is what we are shipping, and what it means for every vendor running a business application.
- [Your coding agent just got bought. Build like you might have to leave.](https://inmydata.ai/blog/your-coding-agent-just-got-bought): SpaceX just bought Cursor. The consolidation at the top is complete. Portability costs less than you think, and coding agents make the discipline that delivers it essentially free. Here is how we build at inmydata, and why this is the right way to work regardless of who owns your tool next.
- [Opus 4.7 has a new parameter called task_budget. Pay attention to it.](https://inmydata.ai/blog/opus-47-task-budget): Anthropic deprecated the parameters developers used to tune how the model thinks, and shipped a single new one for bounding how much thinking it can afford. The direction change is important. This is what it means for what you build next.
- [Security by Obscurity Is Dead](https://inmydata.ai/blog/security-by-obscurity-is-dead): Anthropic refused to release Mythos publicly because of what it could do. The reasons should change how you think about every line of code you ship.
- [Why constraining your agents makes them better](https://inmydata.ai/blog/why-constraining-your-agents-makes-them-better): Token prices have fallen 280x. Enterprise AI bills have tripled. The difference is architecture. Here's what we learned building three agentic systems where every design decision was a cost decision.
- [The knowledge that matters has never been computerised. Until now.](https://inmydata.ai/blog/the-knowledge-that-matters-has-never-been-computerised): For fifty years, we've built systems to store structured data. But eighty percent of what organisations actually know lives outside those systems. LLMs make it possible to computerise that knowledge for the first time. But most approaches are getting it wrong.
- [Hallucination is the feature](https://inmydata.ai/blog/hallucination-is-the-feature): The same mechanism that makes AI dangerous in production is what makes it valuable everywhere else. The difference isn't the model. It's the constraints you put around it.
- [Agent identity is the easy part.](https://inmydata.ai/blog/agent-identity-is-the-easy-part): RSAC 2026 launched a wave of agent identity products. But the pattern that actually makes agents safe and useful already exists in coding agents. Here's how to apply it to enterprise systems.
- [Stop tuning your prompts. Start engineering your context.](https://inmydata.ai/blog/stop-tuning-your-prompts): Every frontier model degrades as you fill the context window. The bottleneck was never capacity. It was curation. Here's what that means for enterprise AI, and what we're doing about it.
- [MCP won. Now build something with it.](https://inmydata.ai/blog/mcp-won-now-build-something-with-it): The protocol war many predicted never happened. MCP is becoming invisible infrastructure, and the real story is what you can build when agents compose multiple types of intelligence through simple, modular servers.
- [The Expert Knowledge Gap Nobody's Talking About](https://inmydata.ai/blog/the-expert-knowledge-gap-nobodys-talking-about): Every company is racing to adopt AI. Almost none of them are capturing the institutional knowledge that makes AI actually useful.
- [AI-First, and Last](https://inmydata.ai/blog/ai-first-and-last): Two real agent systems. Both built by AI, not just powered by it. Why the organisations that use AI to build will outpace those that only build AI.
- [The Problem Isn't the Agent. It's the Approach.](https://inmydata.ai/blog/agentic-coding-discipline): The narrative that agentic coding tools produce poor quality, insecure, hard-to-maintain code is not wrong. It is just aimed at the wrong target. The issue is not the tools. It is how people are using them.
- [How Business Software Survives the Agentic Era](https://inmydata.ai/blog/the-incumbents-dilemma): ERP, CRM and enterprise software vendors are losing billions in market value as agentic AI eats their business logic and UI. Here is what the survivors are doing about it.
- [A Friday Afternoon Pen Test and a Trillion-Dollar Question](https://inmydata.ai/blog/friday-afternoon-pen-test): I built, deployed, and ran a penetration testing suite in an afternoon. That's a perfect case study of why software stocks are in freefall.
- [When AI Eats Software, Who Gets Eaten First?](https://inmydata.ai/blog/when-ai-eats-software-who-gets-eaten-first): February 2026 wiped a trillion dollars off software stocks. That wasn't a bubble bursting. It was disruption arriving faster than anyone expected.
- [The AI Bubble That Isn't](https://inmydata.ai/blog/the-ai-bubble-that-isnt): Why the last six months have made an AI bust vanishingly unlikely, and what the bubble hawks got wrong.
- [We Are Not Doomed to AI Slop](https://inmydata.ai/blog/we-are-not-doomed-to-ai-slop): Slop got a dictionary definition. Here's why it won't need one for long.
- [OpenClaw and the Security Tax of Real Agency](https://inmydata.ai/blog/openclaw-and-the-security-tax-of-real-agency): What a viral lobster teaches us about pent-up demand for personal AI and the price of actually getting things done.
- [Where to Start with AI, A Practical Guide for the Overwhelmed](https://inmydata.ai/blog/where-to-start-with-ai-agents): Every organisation knows they need to do something with AI. Here's how to find your starting point.
- [Why Your Software's UI Is About to Become Irrelevant](https://inmydata.ai/blog/why-your-softwares-ui-is-about-to-become-irrelevant): Software has spent 40 years accumulating complexity. AI agents are about to collapse it.
- [Writing Code Is Becoming the Least Important Part of Software Engineering](https://inmydata.ai/blog/writing-code-is-becoming-the-least-important-part-of-software-engineering): Why Agentic Engineers will become the new standard, Vibe Coders will hit a wall, and Traditional Engineers face a reckoning. A look at what 2025 taught us and what 2026 demands.
- [The Case for Slower AI](https://inmydata.ai/blog/the-case-for-slower-ai): LLMs perform far better when given time to think, plan, and critique their own work. We're only beginning to understand how to work effectively with these tools.
- [Worried About an AI Bubble? You're Asking the Wrong Question.](https://inmydata.ai/blog/worried-about-an-ai-bubble): There is growing nervousness about an AI bubble. But focusing on valuations misses what matters: whether you're positioned to benefit from what AI can already do.
- [The Next Evolution of Intelligence](https://inmydata.ai/blog/the-next-evolution-of-intelligence): Seeing 'humans' in machines - exploring how AI is evolving beyond simple automation toward a new form of intelligence.
- [Building the Foundation for Agentic AI: Why Data Integration Unlocks the Future](https://inmydata.ai/blog/building-the-foundation-for-agentic-ai): Agentic AI is more than the next step in automation - it's the leap from insight to autonomous action. But none of it works without the right data foundation.
- [How Coding Agents Change the Way We Build Software](https://inmydata.ai/blog/how-coding-agents-change-the-way-we-build-software): There's a lot of excitement around the new wave of coding agents, and for good reason. The first generation that works has arrived.
- [Coding with Agents: My Experiences with Vibe Coding](https://inmydata.ai/blog/coding-with-agents-my-experiences-with-vibe-coding): Coding assistants are evolving at remarkable speed. Here's what it's actually like to work with the latest AI coding agents across multiple projects.
- [Making Agentic AI Work in Your Organisation](https://inmydata.ai/blog/making-agentic-ai-work-in-your-organisation): The promise of agentic AI is no longer theoretical. Here's how to bridge the gap between pilot projects and production-ready systems.
- [Things are About to Get Weird: The March of AI](https://inmydata.ai/blog/things-are-about-to-get-weird-the-march-of-ai): If you thought things were moving fast last year, buckle up. The pace of AI innovation is now warping the rest of the software market.
- [Ten Minutes on a Sofa: The Real News Story](https://inmydata.ai/blog/ten-minutes-on-a-sofa-the-real-news-story): While the news obsessed over politics, I implemented a cutting-edge AI improvement across multiple production systems in ten minutes from my sofa.
- [Agentic AI in Action: How Smarter Systems Solve Bigger Problems](https://inmydata.ai/blog/agentic-ai-in-action-how-smarter-systems-solve-bigger-problems): Understanding multi-agent AI systems and the four key patterns for designing cooperation between AI agents.
- [Make Smarter Retail Decisions with inmydata Copilot: Conversational Analytics That Thinks Like You Do](https://inmydata.ai/blog/make-smarter-retail-decisions-with-inmydata-copilot): Discover how inmydata Copilot transforms retail analytics with AI-powered conversational intelligence, enabling faster, smarter business decisions.
- [Why AI Projects Fail (And It's Not the AI)](https://inmydata.ai/blog/why-ai-projects-fail): Most AI projects stall not because of model limitations, but because of data access problems. Here's what we've learned from four years of shipping AI systems.

## Contact

- Website: https://inmydata.ai
- Contact: https://inmydata.ai/contact
- LinkedIn: https://www.linkedin.com/company/datapa-ltd
- X/Twitter: https://x.com/inmydata
- YouTube: https://www.youtube.com/@inmydata
- Address: 14 Albany Street, Edinburgh, EH1 3QB, Scotland, UK