Blog
Notes from the workshop.
How we think about people and agents working together, data that stays where it belongs, and what shipped.
- AgentsBring your own AI subscription: what it means and why it mattersBring your own AI subscription means a tool uses the AI account you already pay for instead of reselling model access. Here is how it works and why.
- PipelinesChange detection: asking "what changed since last time?How change detection works in data: stable keys, snapshots, checksums, separating noise from real change, and turning differences into a report.
- Data custodyData provenance: keeping receipts for every recordData provenance records where each record came from, how it was fetched and whether it has changed. What to store, and why it pays off later.
- PipelinesEntity resolution: working out which records are the same thingHow entity resolution works, from cleaning and blocking to scoring, clustering and human review, and how to handle false and missed matches.
- AgricultureFrost risk explained: radiative and advective frostThe two kinds of frost, why low spots freeze first, what makes a frost night likely, and why the type of frost decides which protection methods can work.
- AgricultureGrowing degree days explainedGrowing degree days measure accumulated heat to estimate crop and pest development. Here is the formula, worked examples, and the limits of the method.
- Data custodyHow this website counts visits without cookiesWe wanted to know which pages are useful without tracking anyone. This site counts visits and clicks with no cookies, no stored IP addresses and no third party.
- PatentsHow to read a patent claimLearn to read a patent claim: preamble, transitional phrase and body, independent and dependent claims, and why every element matters.
- AgentsHuman in the loop: designing approvals people do not skipHuman-in-the-loop approvals only work if people read them. Here is how to decide what needs approval and design prompts that get real attention.
- ProgrammesImpact reporting for funders: numbers you can defendHow to build impact figures a funder can check: clear definitions, counting rules, evidence for each record, honest gaps and careful wording.
- Data custodyLocal-first AI explainedLocal-first AI keeps your files, notes and keys on your own device and sends out only what a task needs. How it works and what it costs.
- TeamsMeeting prep briefs and follow-ups: what good looks likeWhat a good meeting prep brief and follow-up contain, how to write them quickly, and where AI summaries help and where a person still needs to check.
- AgricultureMicroclimate: why the nearest weather station is not your fieldWhy conditions in your field differ from the nearest weather station, which variables differ most, and practical ways to get closer to field-level weather.
- Data custodyOn-device meeting transcription: how it works and trade-offsOn-device transcription turns speech into text on your own computer, so meeting audio never leaves it. How it works, and what you give up.
- TeamsOne workspace, any AI, your data: why Prism Labs existsAI can do real work now, yet most organisations cannot hand it the work: scattered tools, data that cannot leave, answers nobody can defend.
- PatentsPrior art search basics for inventorsWhat prior art is, how novelty and obviousness work in general terms, and how to search patents by keyword, classification and citation.
- AgentsRunning more than one AI coding agent: why and howRunning several AI coding agents lets you work in parallel and cross-check results. Here is when it helps, how to set it up and what to watch for.
- AgentsSkills, plugins and tool servers (MCP) in plain EnglishSkills teach an agent how to do a task, tool servers give it new abilities through MCP, and plugins bundle them. Here is how each works and the risks.
- AgricultureSoil moisture sensors explainedThe two families of soil moisture sensor, volumetric and tension, how each works, what field capacity and wilting point mean, and how to read the data.
- AgricultureSpray windows explained: wind, humidity and inversionsA spray window is a period when weather lets a spray reach its target. How wind, temperature, humidity, Delta T and inversions affect spray drift.
- TeamsToo many work tools: what consolidation really takesWhy adding work apps makes the day harder, what real consolidation involves beyond cancelling subscriptions, and a practical way to audit your tools.
- ProgrammesTracking startup outcomes after an accelerator or programmeHow accelerators and university programmes can track what happens to startup teams afterwards: definitions, evidence, cadence and honest gaps.
- Data custodyTwelve questions to ask an AI vendor about your dataTwelve plain questions to put to any AI vendor about where your data goes, who can read it, how long it is kept and how you get it back.
- AgricultureWhat a digital twin of a farm looks likeA farm digital twin is a living model of fields, soil, weather, crops and machines. Here are its layers, how it differs from a map, and how to start.
- Data custodyWhat actually gets sent to an AI provider when you ask?A question to an AI model carries more than the words you typed. Here is what a request usually contains, and how to keep it small.
- PipelinesWhat is a data pipeline? Seven steps from raw records to answersA plain-English guide to data pipelines, from gathering raw records and keeping receipts to linking, review, sourced answers and delivery.
- PipelinesWhat is a digital twin?A digital twin is a digital copy of a specific real thing, kept current with data from it. Learn the parts, the uses and the common pitfalls.
- AgentsWhat is an AI agent workspace?An AI agent workspace is one place where people and AI agents share conversations, files, tools and approvals. Here is what it contains and why.
- AgentsWhat is an AI agent, and how is it different from a chatbot?An AI agent is a language model that can take actions with tools in a loop, not just reply. Here is how that differs from a chatbot and why it matters.
- Data custodyWhat is data custody in AI?Data custody in AI means knowing who holds your data, where it sits, who can read it and how you get it back. A plain guide to the idea.
- AgentsWhen agents owned by different people work in one roomWhen several people bring their own AI agents into one shared space, ownership, access and approval need clear rules. Here is a practical way to set them.
- Data custodyWhere your AI keys and logins should liveAI keys and logins belong in your operating system's keychain or a secrets manager, not in code, chat or a vendor's database. A practical guide.
- Data custodyWhy AI answers need sources: grounding and citations explainedLanguage models write fluent text whether or not it is true. Grounding and citations tie each claim to evidence you can check. Here is how.
- AgentsWhy coding agents should work on a copy of your repoA coding agent should work on an isolated copy of your repository so you can review one diff before anything merges. Here is how git worktrees help.
- ReleasesPrism Desktop 0.5.33: every agent gets its own setupEach agent now picks its own model and account, with a private configuration of its own. Your native command-line logins are left untouched.
- ReleasesPrism Desktop 0.5.30: personal logins stay personalYour personal Claude Code login stays on your device even inside an organisation, and the desktop app now talks only to the public Prism API.
- ReleasesPrism Desktop 0.5.31: a truthful agent runtime checkPrism now recognises its own bundled Codex bridge correctly, restores the ready state and model list, and checks all thirteen agent programs it supports.
- ReleasesPrism Desktop 0.5.32: agents that reply reliablyA reliability release. A brief connection problem no longer makes a Codex agent abandon its turn or answer the same message again and again.
- ReleasesThe first public Linux builds of Prism DesktopPrism Desktop's first public Linux builds went out in early August, with signed automatic updates. Here is what that first month of releases covered.