Andy Cotgreave builds interactive dashboard with Build Agent

Thanks Andy Cotgreave for this thoughtful review! We built Ridge AI to help people build robust, shareable analytics, not just pretty dashboard slop. You called out some of the design elements that we use to do that. Re: Tweaking: the underlying charting language, Ridge Plot, is quite rich and expressive. Ask the Build Agent to change something and you'll be surprised how much it can do. 😃

My current favorite AI task is creating mockups. Our team has spent countless hours over the years creating samples to illustrate what we envision building (a data architecture drawing, a dataset, a dashboard, a scorecard, an AI interface etc etc). It was enormously time consuming and often failed paled in comparison to the full vision of what we had in mind. And I struggled with the reality that mockups require time and work, and are actually the first phase of a scope of work. But they have to be done before a penny is paid in order to illustrate our outcomes and deliverables. And sometimes we don't win the work, losing time, energy, and money. But now AI can do the mockup work in minutes, including fabricating sample data and generating illustrative visuals, often showcasing our vision better than we could manually. Clients are using it too, to show us what they want next. Enormously efficient for them and for us. The human expertise and judgement in forming an intelligent scope of work doesn't change. But now the picture of it doesn't cost either of us a week.

The hardest part of building a dashboard isn't adding more data. It's figuring out what doesn't need to be there. I've been spending a lot of time on the UX for ‏XORV‏ lately, and AI search data gets complicated pretty quickly. Different models, citations, competitors, time ranges… there's a lot you could put on one screen. And honestly, most of it makes perfect sense to me. I built it. Of course it does. 😅 But someone opening the product for the first time probably just wants to know: Are we showing up? What's changing? What should we do next? That's what I've been trying to design around. Funny how writing more code is sometimes easier than deciding what to leave out.

Quick question: when was the last time you opened a dashboard because you *wanted* to? For me, it's usually because someone asked me something and I knew the answer was in there somewhere. A few clicks, a couple of filters, and an export to Excel later, I've got what I need. A coworker and I got into the topic of dashboards recently. He's doing discovery work on a sales enablement dashboard, and the bigger goal is improving the overall user experience. Somewhere in that conversation I said what I'd been thinking for a while: I feel like the way we think about dashboards is going to change significantly soon. The data still matters - maybe more than ever. But I think the future looks less like a screen full of charts and graphs that some brilliant UX designer curated for you, and more like a well-governed data source you can just ask: "Which of our deals stalled this quarter, and what information could I send to help get them moving again?" And I'm sure AI could put together a set of graphs if you really need it for a presentation or something. What I'm less sure about is whether or not users will trust this method of accessing the data they need to work more efficiently. A dashboard that's human built and validated feels safer than a robotic, AI-generated response. But maybe that's just because it's familiar? So I'm curious: Are you still investing in new dashboards? Has anyone actually replaced a traditional dashboard with AI-driven querying? How'd it go if you did? As a user, do you want to build dashboards or just interrogate your databases by asking natural language questions?

OpenRouter 101: Add models to Codex, Claude Code In this guide, you will learn how to connect OpenRouter to your AI agent so you can access hundreds of models through one account, compare pricing, and test cheaper options for tasks like ad mockups. Note: Connecting Claude Code to OpenRouter routes your requests and billing through OpenRouter credits instead of your Anthropic plan. You do not need an Anthropic subscription for this, but you are now paying per OpenRouter credit, not a flat monthly fee. Step-by-step: 1️⃣ Sign in to OpenRouter and browse the model catalog. Filter for your task type, sort by price low to high, and open a model page to check its capabilities, context window, and cost per token. Why this works: Comparing models side by side before committing means you are not stuck defaulting to whichever model your agent uses out of the box, which is rarely the cheapest or the best fit for a specific task. ⸻ 2️⃣ Go to Keys in your OpenRouter dashboard and click Create Key. Name it after the project or agent you will use it with, set a spending limit, and set an expiration date before generating it. Copy the key immediately. It starts with sk-or-v1- and is shown only once. Why this works: A spending limit and expiration date protect you from a runaway bill if a script misfires or a key gets exposed accidentally. ⸻ 3️⃣ To connect Claude Code, add these three environment variables to your shell profile: export ANTHROPIC_BASE_URL="https://openrouter.ai/api", export ANTHROPIC_AUTH_TOKEN="your-openrouter-key", and export ANTHROPIC_API_KEY="". The last one must be explicitly empty or Claude Code can fall back to billing your Anthropic account directly. Why this works: The empty ANTHROPIC_API_KEY line is the single most common setup mistake. Skipping it means you could be billed by both Anthropic and OpenRouter for the same session without realizing it. ⸻ 4️⃣ Give your agent a small, low-stakes task using a cheaper model first. For example: "Use a lower-cost image model on OpenRouter to generate three ad mockup variations for [product]." Compare the output quality against your usual model and check the actual cost in your OpenRouter usage dashboard. Why this works: Testing on a low-stakes task first shows you exactly where a cheaper model holds up and where it falls short, which is more useful than trusting a price tag alone. ⸻ 📌 Save this. Send it to any developer or builder paying full price for one model when a fraction-of-the-cost model would do the job just as well for routine tasks. ⸻ 💡 Pro Tip: Ask your agent to create a reusable skill using OpenRouter for cheap, bulk variations whenever you are brainstorming. Generate ten rough options at a fraction of the cost, then send only the winning direction to your primary model for a polished final version. ⸻ Follow BTEHub for daily AI productivity tutorials built for builders. https://lnkd.in/d5s4S6J5 BTEHub Team

Your agent can read the dashboard. But can it do the paperwork? Reading is where most agents stop. Ask one to create a record, fill out a form, or update a field — and it's back to reading pixels: find the input, guess the label, hope the layout didn't change. Move one field, rename one label, and the whole workflow breaks silently. You find out from a customer. HIVE treats forms and records as structured actions instead. A request comes in plain English; the agent gets the form's fields as typed actions — names, types, what's required — and fills and submits. Create, read, update: each one a declared action with a declared result. You see exactly what changed, and the agent sees it too. Structured actions don't care what the UI looks like. No pixel-hunting, no layout guesses, no quiet breakage when someone redesigns the page. And write access only goes as far as your permission rules allow. Destructive actions wait for a human yes. Demo video #2 covers the whole flow — form fill, then create, read, and update on a record — all without the agent ever looking at the screen. What's the form in your product an agent would break on first? #AIAgents #AgenticAI #BuildInPublic #DevCommunity #SoftwareEngineering #AI

📺 Nobody wants to watch 100 session replays 😵💫 🤯 That’s always been the problem - watching the right sessions can be incredibly revealing. But when you have thousands of sessions backed up, someone needs to sit through them to find that 🪡 in a haystack. So we’ve built something to do that work for you. I'm super-pleased to share that Zuko can now use AI to "watch" and analyse your form session replays and identify what’s causing people to struggle or abandon. Rather than spending hours watching individual sessions, you automatically get a summary of the patterns across them: ➡️ The main reasons people are abandoning ➡️ Likely explanations for that behaviour ➡️ Other friction points worth investigating ➡️ The sessions that provide the evidence You can then jump into the relevant replays yourself to see exactly what happened. And, as it's melded with Zuko data, it automatically knows where the most pressing issues are so it starts there. Hopefully this is something where AI is genuinely useful - saving you hours that can be used productively elsewhere. Do you have any session replay hero or horror stories? Share them in the comments 👇 _____________________________________________ 👋 I’m Alun - MD of Zuko Analytics; in-depth form and checkout analytics software. I post mainly about form UX, CRO and optimization with a bit of AI thrown in for good luck.

💡 What if one analytical standard could power two different dashboard experiences? My latest project, No-Code Dashboard Delivery Framework, explores exactly that: AI-assisted Excel for maintainable analytical ownership and AI-generated HTML for portable leadership and client consumption. What I valued most was designing beyond the visuals. The framework combines data integrity, evidence-based KPIs, decision storytelling, refresh validation, accessibility, and reconciliation into one reusable approach. For the HTML path, the workflow moves from data discovery and field mapping to model building, dashboard composition, testing, reconciliation, and packaging into a portable browser dashboard. For me, the takeaway is simple: AI can accelerate dashboard development. Strong analytical discipline is what makes the output trustworthy. #DataAnalytics #AI #M365Copilot #Excel #HTML #DashboardDesign #DataVisualization #Innovationv

Prompt to interactive web dashboard in minutes with Kimi AI. ⚡ Turned a finance tracking concept into a live, interactive web dashboard in minutes using Kimi AI! All it took was generating the full frontend code, saving it locally, and launching it straight in Google Chrome. Seeing clean UI components and dynamic charts come to life instantly felt like magic. What the FinanceFlow dashboard includes: Executive Metric Cards: Real-time tracking for Total Income, Total Expenses, and Net Savings Balance. Monthly Savings Trend: Smooth area/line visualization capturing cash flow shifts across the year. Expense Breakdown: Clean donut chart categorizing housing, food, transport, and subscriptions. Transaction Flow: Quick interactive transaction entry form paired with a structured recent activity ledger. AI web prototyping is completely changing how fast we can turn data concepts into deployable interfaces. What AI tools are currently speeding up your analytics and UI workflow? Let’s connect in the comments! #DataAnalytics #KimiAI #WebDevelopment #DashboardDesign #DataVisualization #FinanceApp #ArtificialIntelligence

FRAMEWORK THINKING FOR AI: FROM AD‑HOC PATTERNS TO REUSABLE DESIGN Early AI projects often rely on ad‑hoc patterns: a clever prompt here, a custom integration there, a one‑off evaluation script. This is fine for exploration but fragile for scale. Framework thinking means identifying recurring problems—retrieval, tool orchestration, evaluation, governance—and designing reusable patterns that teams can apply consistently across use cases. Conceptually, a framework is more than a library; it is a way of thinking about problems. For AI, that might include standard prompt structures (roles, constraints, outputs), orchestration flows (input validation, model calls, tool routing, post‑processing), and evaluation loops (offline tests, online monitoring, incident response). When these patterns are documented and supported by code, they become a shared language across teams. The misconception is that frameworks reduce flexibility. In reality, they reduce accidental complexity. Teams still have room to innovate within the framework—choosing models, crafting domain‑specific prompts, and designing UX—but they do not have to reinvent core mechanics like trust boundaries or logging. This improves both velocity and safety. From a system‑level perspective, framework thinking leads to layered architecture: core services for capabilities, libraries for common patterns, and templates for specific domains. Governance and platform teams can embed policies into the framework, ensuring that every new use case inherits baseline controls. Over time, the framework evolves with the organization, capturing lessons from incidents, evaluations, and successful deployments. #AI #Frameworks #Architecture #LLMApps #PlatformEngineering #Reuse