Beau Cameron

Every team builds their own workflow engine. Most regret it. Time to see if the purpose-built solutions are actually better. As a follow up to my Enterprise Integration Patterns blog series, I'm diving into workflow orchestration systems. Throughout the posts so many folks opined "I feel like I build these all the time, but nothing solves it in a general way" and I'm interested in finding that general way. I'm planning to evaluate: - Temporal - Amazon Simple Workflow - Vercel Workflows But I know this space is vast. What workflow orchestration systems are you using in production? And more importantly: what questions would you want answered in an honest evaluation? Things like: developer experience, failure handling, observability, scaling characteristics, or something else entirely? Drop your suggestions in the comments. Your input will directly shape what I explore over the coming weeks!

5

4 Comments

🤔 I've been thinking a lot about technical debt lately...it's not boring I promise! More specifically, about how we all *deal* with technical debt. Over the years, I - like many others - have pretended it didn't exist, or focused only on the quick wins. What I hadn't really considered until recently, is that sometimes we take on technical debt consciously - and that's okay. Sometimes it's the right call. Sometimes it buys learning. Sometimes it buys time. The real problems don't usually come from the debt itself, but from the lack of visibility around it - when no one remembers why it was taken on, or when it should be addressed. Maybe it's time we stop avoiding technical debt and start talking about it openly - so it can be managed properly. How do you think about technical debt on your teams?

22

13 Comments

Power Pages quietly crossed an important line. With GA support for single-page applications, developers can now bring React, Angular, or Vue apps into the platform using the same tooling and workflows they already use. But this isn’t just “SPA support.” You can pair it with any AI tool of your choice to generate, refactor, and iterate on the front end — while Microsoft handles identity, data access, security, and hosting underneath. So instead of: • choosing between a platform or a framework • rebuilding auth and APIs for every app You get: • framework freedom • AI-accelerated development • enterprise guardrails by default And most importantly — choice. Prefer traditional Power Pages for fast, low-code delivery? Use that. Want full pro-code control with a modern SPA stack? Use that. Same platform. Same security. Multiple choices on how you build. This isn’t low-code becoming “more pro-code.” It’s adapting to how modern developers actually build. https://lnkd.in/gDXudMkY #PowerPages #BringYourOwnCode #CodeSites #PowerPlatform #BYOC

98

3 Comments

MERGE is GA in Fabric since January 2026? Stop the presses. The cynical voice in my head can’t help but note: ANSI SQL standard since 2003. SQL Server had it in 2008. But that’s not the interesting part. Is MERGE really that big of a deal? Yes being able to perform INSERT, UPDATE, and DELETE in a single atomic unit is useful. No debate there. But if your engine already supports proper transaction isolation, you were never exactly blocked. Let’s be honest: writing a MERGE statement is rarely anyone’s idea of fun. And no two platforms implement it quite the same way. Now here’s the more interesting angle. In cloud analytics platforms built on immutable storage, what does UPDATE or DELETE actually mean? Immutable storage + row-level mutation is a strange combination: - On Snowflake, merging VARIANT data with 90-day Time Travel enabled: have you looked at the size of Fail-safe lately? - On Databricks, when you run large merges, how much data does VACUUM actually reclaim? - On Fabric, does OneLake ever feel… fuller than expected? Or is storage simply abstracted away from the bill? These systems don’t really “update” rows. They rewrite files. Storage is cheaper than it used to be. It is not free. And keeping old data around has financial, operational, and sometimes regulatory consequences. Personally, I prefer CTAS-style patterns over MERGE, regardless of syntax. CTAS is simple. Explicit. Predictable. It aligns naturally with immutable storage instead of simulating mutation on top of it. It also gives you the flexibility to materialize as a table or a view at the eleventh hour. Is CTAS always the right answer? Of course not. Good engineering is not about absolutes. It’s about understanding trade-offs. So this isn’t about celebrating or mocking Fabric’s late arrival to the MERGE party. It’s about a more fundamental question: To merge or not to merge?

12

Azure discounts are one of those things that feel obvious until you actually look at a bill. After writing about Azure Savings Plans and why the savings don’t show up the way people expect, the next most common question is: “What happens when Reserved Instances are also in play?” This post tackles that exact moment of confusion. It walks through how Azure actually decides which discount applies, why discounts don’t stack, and why a bill can look messy even when everything is working exactly as designed. If you’ve ever looked at cost reports and thought “this should look different,” this might help reconnect the dots: https://lnkd.in/gxa7AnRm #Azure #CloudCosts #SavingsPlans #ReservedInstances #FinOps

14

1 Comment

Claude Code now has built-in LSP support. This will produce more consistent code changes, allowing for Spec Coding of larger projects. It is an important upgrade! Read about what that means here: https://lnkd.in/eThKmKAd

6

Quick tip for anyone building custom Typst templates for use with Quarto: If special characters like $ are causing mysterious compilation errors that standard escaping won't fix, try hexadecimal escape sequences instead. For example: \u{24} for the dollar sign. Wrote up a quick note covering this technique, including how to use Typst's ASCII table tool to find hex codes for any character. Could save you some debugging headaches! https://lnkd.in/eE62SBGb #Typst #Quarto

39

A thing I've been hollering about this year are the Design Guidelines everyone should think about when building AI Agents. If you are using M365 Agent Builder or Copilot Studio, these are already built-in guardrails (so they're a good place to "practice") before you go full-on Agent-A-Saurus-Rex 🦖 Below are a set of guidelines that we released in February that were centrally drafted by Office of the CTO (OCTO). If you look closely at these guidelines, you will notice that in general, these are just "good design" rather than getting overly creative with your bad self and inventing new things that confuse devs and scare away customers. Have a look and holler at me with questions! 👀Transparency •Let the user know this is an AI-enabled Agent (eg, this is an AI Agent, please validate responses, etc) •Provide basic instructions to get started (e.g. a “Hello” message, sample prompts) •Show history of prompts that the user asked in the past •Make it clear how to give feedback (e.g. thumbs up & down, "Send feedback" button, etc.) •Clearly articulate any usage or topic restrictions the agent has (eg, this agent only talks about travel, etc) 🚘Control •Describe how the dev or user can modify the Agent, after it’s been created, with things like the System Prompt •Enable the dev or user to choose the Agent's comms style: how verbose it is, desired formats, any caveats on what the Agent should not talk about. •Allow the dev or user to view and delete history: any associated files or data, prompts, and past conversations. 📎Consistency •Make sure icons are standard and recognizable (e.g. "Share Prompt," "Add a file or photo," "tag someone") •Use paperclip icon to indicate file upload/sharing with the Agent, and an image icon to indicate graphics upload.

115

11 Comments

You can leave Percona, but you can't escape the Percona mindset. 😊 When I saw the news that Drupal is dropping MySQL (and Percona) as a recommended DB, I had to investigate. I traced the issue back to a bug regarding Percona XtraDB Cluster submitted almost six years ago. (https://lnkd.in/dwxfsHNN) What I found was frustrating. For six years, rather than submitting a bug report to Percona or asking for clarity, the maintainers kept patching the code based on assumptions - like the idea that AWS Aurora and XtraDB are functionally the same. This is a failure of collaboration. Open Source is not just about publishing code; it's about the ability to work together across organizational lines. - See a bug? Ask the maintainer. - Not an expert? Find one. - Patching the same issue for years? Get help. Let's get back to the core principles of our industry. Discuss your problems, even with people who work for your competitors. P.S. Help the community protect that ecosystem. Sign Vadim's Open Letter to Oracle here: https://lnkd.in/dx3geV6Z

6

CRUD isn't enough. Agree, it usually isn't nearly enough for anything functional and usable. That's why almost all app generators that can build an entire app in a few clicks fail, and they fail miserably because they all generate code (endpoints, validation, even the UI) around CRUD operations. Real-world applications need more than one table; they need joins, transactions, and all that stuff.

34

12 Comments

Tech debt doesn’t show up on your P&L. But it does show up in delays, bugs, and mounting developer hours. If you're running a Adobe Commerce store, you've probably experienced it firsthand... - Custom features that break with upgrades - One-off code only one developer understands - Fixes that fix the fix from last quarter That’s tech debt. It builds up over time and quietly erodes your margins. After 15 years building on Magento, we’ve learned that the key to reducing tech debt isn’t just cleaner code. It’s a shift in mindset. At Zero-1 we’ve built a more scalable, multi-use approach to development. Think of it as a SaaS mentality...while still retaining complete control over your platform. In concrete terms, this means: - More reusable code - More standardized extensions - Less time reinventing the wheel The result is lower long-term costs and a store that’s easier to grow. If you're planning a new feature or re-platforming project and want to build with the future in mind, let's talk!

6

Openclaw is great for tracking and scheduling home maintenance items. Take for example car maintenance: 1. Agent script runs and syncs mileage from Home Assistant 2. Calculates remaining miles for each scheduled maintenance item based on intervals stored in Sqllite DB 3. Checks for any Trello cards you moved to "🎉 Done" → marks maintenance complete at current mileage, calculates next due date 4. Creates new Trello cards in "To-Do" for any items that are due and don't already have an open card Workflow: 1. Get notified via Telegram when maintenance is due 2. Find the ticket in Trello "To-Do" list 3. Do the actual work (or schedule it) 4. Move the Trello card to "🎉 Done" when complete 5. Next day at 9am, script sees the completed card, records the mileage, and won't bug you again until the next interval Example: • Tire rotation ticket created (overdue at 32,230 mi) • You get it done, move card to Done • Next morning: script sees Done card, records "completed at 32,230 mi", next due at 47,230 mi • No new ticket until you hit ~47k miles Pretty cool, right?

6

We are soon shipping a lot of QoL improvements to the Business Central Word add-in for document reporting. All related to UX for the layout author. More updates are coming over the next weeks. You can try it today by installing the preview version. Note that Word does a lot of caching, so you might get an install error the first time you try to install the add-in. If so, remove any preview add-ins (for Business Central), wait a day or two, and then try again. #msdyn365bc #documentReporting #word

74

3 Comments

Most vibe coders are optimizing the wrong layer. There’s a post going around showing how a “non-technical” prompt and a “senior engineer” prompt produce wildly different results for the same task. The takeaway people grabbed: write longer, more detailed prompts. That’s not the lesson. Here’s what usually happens. A deal team needs to review 6,000 documents in a virtual data room for a mid-market acquisition. Someone asks an AI to flag every contract with a change-of-control provision. They get a script that searches for the phrase, pulls matches, and dumps them into a spreadsheet. Half the results are boilerplate. Key risks in assignment clauses and consent requirements get missed entirely. Nobody knows what the AI skipped or why. An experienced builder doesn’t write a better search query. They ask different questions first. What document types are actually in this data room? Are we looking for explicit change-of-control language, or any provision that could block or complicate closing? Do amendments override the base agreement? Are side letters in the set? What’s the priority: speed to first-pass flags, or completeness for the disclosure schedule? What happens when a clause is ambiguous and needs deal-team judgment? That reframes the problem before a single line of code gets written. The gap between a weak AI output and strong output isn’t verbosity or technical vocabulary. It’s the quality of the constraints. Success criteria. Tradeoff decisions. Knowing what not to build. Vibe coding gets dramatically better when you stop describing the task and start defining the boundaries around it. I Try this: next time you prompt an AI to work through a diligence set, write three sentences before the request. What’s a real risk versus noise? What should get escalated instead of auto-flagged? What’s the consequence of a false negative? That alone will change the output more than any prompt template. Subscribe to PossibLaw. Substack. Podcast. Tools to become a builder. We run custom training for legal teams who are done watching from the sidelines. Let’s talk.

2

2 Comments

If you feel left out of Claude Code cause you aren’t technical Download Claude Desktop It has Chat | Cowork | Code side-by-side Chat you already know how to do Cowork is a mini-OpenClaw without the hazards And Code is right there, ready to use. All you need to do is ask Chat how to get started with it! That’s how I went from being completely intimidated to running 6 code projects simultaneously in three weeks and looking like an 80s tv show hacker 🖥️

3

1 Comment

For any non-devs using Lovable, they just released one of the most helpful features I've seen... ability to mention code files in chat!! From their change log: https://lnkd.in/ekGjXfUi "Reference specific files directly in chat by typing @ or clicking the reference button in the code editor. This makes targeted edits and discussions about specific files faster and clearer." This has been a game-changer for me! Step 1: type @ Step 2: Lovable suggests files. Pick one or keep typing. Step 3: Select the file you wish to update Step 4: Code file updated 😄 Eazy breezy!

2

The Bureau of Economic Analysis now has a custom connector. The BEA API doesn’t behave like most. You have to get a little weird with it. Read Interface, the latest edition of Developments, to learn how to bring in vital economic data via Power Platform. Article: https://lnkd.in/euKSGYM8 GitHub: https://lnkd.in/ed4VupNG

2

Week 2 of Spring into AI challenged us to write an tool that AI can use. I decided to create a SQL Server DB Profiler skill that an agent can invoke. I wanted to move metrics closer, within my IDE. Usage: "DTUs comin in hot! Check last 3 hours. Show top 3 offenders." Output: A synopsis of top N offending queries over last N hours, invocation counts, cpu waits, and tuning recommendations. I chatted with GPT about what kind of metrics are possible to query SQL Server for. Then I asked it to produce a small suite of sql scripts and a powershell script to map input sentiment to input params in the sql scripts, and aggregate the output into a synopsis that relays the real current state of things, so I can quickly take action. Great for when those monitoring alarms go off.. I mean, of course, they never do, but just in case ;) src: https://lnkd.in/gnRDjsNd https://lnkd.in/gFPY2sAn

9