John Smith

If you follow me on Twitter, you might remember a post where I mentioned solving a tricky architectural challenge while working on the official PHP MCP SDK and how PHP Fibers came to the rescue. We needed a way to pause execution mid-request, let the server talk to the client, wait for a response (in some cases), and then resume exactly where we left off, all while keeping the developer API looking standard, synchronous and simple. I promised a deep dive into the solution, and today I’m delivering it. (🔗 Link to the article is in the comments) This is probably the longest and most comprehensive article I have ever written. It’s not a quick tutorial. It’s technical, thorough and designed to reward patient reading. I went all in discussing the exact problem faced, how Fibers solved the issue for MCP’s client communication flow, the full architectural patterns used for cooperative multitasking in the SDK, and a build-from-scratch example of a cooperative task manager. It’s a lot, but I promise, you will walk away with a solid understanding of how to leverage Fibers for your own complex systems when the need arises. Cheers 🥂

29

1 Comment

Month 1 selling to developers: 'How hard can it be to understand Android development?' Month 6: Can now distinguish between Java and Kotlin in conversation Month 12: Actually understands why everyone gets excited about Compose Month 18: Accidentally used the term 'refactor' correctly in a meeting Month 24: Caught myself getting genuinely excited about backward compatibility improvements Turns out developers LOVE talking about their craft when you ask genuine questions.

19

4 Comments

It’s Sunday. I’m grafting. [I don't do this every Sunday, sometimes I sometimes take time off work] With Mobtag Asset Management, you scan a QR code and complete a form. Simple. Now imagine this… You fill in a form with certain values. The moment it’s saved, it evaluates a set of rules. If the conditions match → it fires the actions you’ve defined. Right now, that action might be something like “Update asset values.” But that’s just the beginning. Because a rule engine exists (design patternz and all that), the action can be anything: - Email someone. - Text someone. - Create a new asset. - Change a status. - Assign a user. - Trigger a workflow. All I need to do is create a new Action class and plug it into a config file. Scan → Submit → Evaluate → Act. It’s about making forms smarter, removing business pain points, and quietly saving time and money in the background, without anyone needing to think about it. e.g - An engineer scans a boiler, fills in a maintenance form. If pressure > threshold OR inspection_status = fail -- THEN -- - Auto-create a follow-up job - Email facilities manager - Change asset status to “Unsafe” - Lock booking system for that asset 🙂 P.S. And it's got tonnes of tests.

9

1 Comment

Assumptions can cause so much trouble and the assumptions that we don't know we're making are often the worst ones. Sometimes, the trickiest thing as a software developer is understanding what people want and need. We don't realise how much we make assumptions when explaining something, or when hearing/reading someone else's explanation. Miscommunication is common, but in most cases it isn't so important, or it will be resolved later on in the process, usually without thinking about it, because people are generally fairly good at using common sense and asking questions when necessary. Building code around incorrect assumptions can produce incorrect results though, and computers don't realise when an assumption is wrong - they just keep doing what they are instructed to. Even AI (ie. LLMs) will rarely notice incorrect assumptions. It is very easy for an assumption to slip in in a prompt that implies something that isn't what the user intended, or for the AI to "make an assumption" in the response which is incorrect and for it to get missed by the user. As we rely on AI more, there is a tendency to trust it more and I think that this becomes a greater problem. Most people aren't actually very good at spotting this, or realising that is why they are getting bad results. I'm not sure if this is something that can be "improved" or whether this is a fundamental problem with AI (or at least the current approach). It doesn't make it useless, but it is something that we need to be conscious of when building AI-based solutions. On a small scale, if the user is using the AI as an assistant and paying attention, this is usually caught fairly quickly. When AIs are left to work more independently as "agents", doing a lot more for themselves, the risk of these "assumptions" snowballing and producing something significantly different to what the user intended starts to become a much bigger problem. What's worse is that larger pieces of work like this are much more tedious to check for users and can be hard to plough through. That said, I have found some uses for AI working at these scales, but I personally only use it in limited cases at the moment. In either case, most guidance on developing AI based solutions recommend keeping a "human in the loop" as best practice, partly to ensure that the AI doesn't go off the rails without anyone noticing. We have decades of experience of building business systems. If you are interested in developing a system for your business (with or without AI), get in touch - I'm happy to have a chat about it.

6

People keep telling me that I should be enjoying AI for software development and apparently find it hard to understand why I don't. Putting aside the many reasons why AI is problematic, including the parallels to addiction-forming behaviours, I think I can finally explain what part I really have a problem with. It might even be classifiable as a skill issue, but if so it's not one I can readily solve, nor do I even wish to. You see, I am in the camp of being somewhere on the autism spectrum. I don't think that should be a surprise to anyone. But how it manifests might be. I do not, for example, have any intuition whatsoever with body language. Whatever I pick up is a conscious thought effort on my part after lots of practice and misreads. Human comms is the worst, though. There is what people say, what people mean, what they think they mean and what they think I want to hear about any of it. And sifting between those when you have no real intuition for it is an interesting problem. I tend to default to taking people at their word because it is really the only concrete thing I have, everything else is advanced guesswork. Computers by comparison have none of this. Everything is 'do what I say'. Not 'do what I mean', not 'do what you think I mean'. It's all firmly 'do what I say'. There's no ambiguity, no second guessing. Talking to an LLM is very akin to the worst elements of talking to a human; it's so biased towards 'do what you think I mean' or 'say what you think I want to hear' (as crude anthropomorphic simplifications) that trying to navigate that is more stressful than talking to actual humans. Especially with the fake "you're absolutely right!" nonsense. I can give 100% effort for 100% of a unit of work in 1 unit of time... Or I can use an LLM, put in 150% of the relative effort for 120% of a unit of work in 1 unit of time. (And this is the figure hundreds of hours in. It was closer to 200% effort for 105% of baseline output.) Did it accelerate me? Sure, compared to the baseline - I got more done in the same. But it cost me more effort to do, in an unenjoyable way, and I don't think any amount of skill is *ever* going to fix that. Not unless you can rewire my brain. But then we get into the fact that I as a person don't need fixing. This might not be true for you. But you're not me, and you're probably not where I am on the spectrum.

270

96 Comments

"In a world where people are digitally literate without being technical, the software that succeeds will be the software that feels as though it already knows what you came to do." I wrote about this for today's Yorkshire Post For 20 years, business software has been built on the assumption that users need to be taught how to use it. Sidebars, burger menus, dashboards showing everything at once. I think that assumption is dead. Non-technical professionals use AI daily, navigate complex tools instinctively, and don't need a sidebar to find settings. The muscle memory has shifted from clicking through structure to expressing intent. But most enterprise software hasn't caught up. And here's what I think most people are missing. The next generation of users won't always be human. AI agents acting on behalf of people need the same thing: context in, outcome out. No navigation to learn. No hierarchy to click through. Software built around intent is agent-ready by default. Software built around sidebars isn't. That's what we're building at AUDITSU. No sidebar. No traditional navigation. No dashboards overloaded with metrics nobody asked for. One ambient surface that compresses interface complexity into contextual intelligence... "Ambient Intelligence". You focus the system. It responds. Whether you're a compliance officer or an agent working on their behalf. When a system understands context and adapts in real time, users stop navigating toward what they need. It just appears. Full article linked in the comments.

33

6 Comments

When you’re asked to lead a project in a framework you’ve never used… 🛑 Challenge My first role as a Tech Lead—and I joined a team already two weeks into an Angular project. I was a React developer by training, completely new to Angular, yet I was asked to guide those two engineers and own delivery. 💡 Action • Embraced the learning curve: Dove into Angular docs, tutorials, and sample apps. • Paired programming: Ran daily sessions to onboard my teammates and share emerging best practices. • Translated patterns: Adapted familiar React and TypeScript patterns into Angular’s structure. ⚙️ Results • The client loved our momentum and awarded us a follow‑on project. • Our billable team grew from 3 → 8 engineers in just a few months. • I received a significant appraisal hike for taking on the challenge—and delivering. 🔑 Takeaway Learning new libraries, frameworks, or technologies makes me a stronger, more adaptable engineer—and it invariably earns recognition. 🤔 Your Turn Have you ever stepped into a new technology and come out stronger? #LeadershipInTech #TechLead #CrossSkilling #FullStack #CareerGrowth #Learning

15

Software engineers looking to break into the world of VC-backed, early stage & hyper growth environments, please read on. In the last month, I've had 554 engineers apply to only 3 adverts. Most of these engineers weren't suitable - not necessarily because of skill, desire or intention - for two reasons: 💡 They hadn't positioned themselves as early stage operators. 💡 They didn't understand the expectations and what founders look for. I've repeated myself countless times recently with people who found the feedback helpful, so here's the plan. The PROBLEMS ❌ VC-backed founders are fighting over the same micro talent pool. ❌ Engineers want to break into this world but most don't know how. The MARKET 🌍 Corporations are laying off by the thousands because of AI 🌍 People are awake to the fact they can make more impact in start ups The SOLUTION 🔥 Engineers need to know founder green & red flags 🔥 Engineers need to know the personality traits that succeed So I'm running an exclusive webinar in April with limited spaces. This is for serious people who want to break into this world but can't. This is education and awareness, NOT a guaranteed job. Only apply if you're serious about levelling up. But why trust me? I'm recruiting for companies backed by a16z, Sequoia, Lightspeed and YC. I'm in slack channels with ex-Google AI Engineers, top 0.01% physicist, international chess master and multi-exited founders who've raised well over $100m collectively. I hear the same things, day in, day out. And I'm sharing it all with you next month. Link in comments.

30

2 Comments

Queues and JavaScript​: Do We Need Queues at All? Apparently not, with a little bit of magic (dark magic) and determination, we can implement a solution that uses arrays as queues, without any performance implications (but I wouldn't recommend using it in production; it was just for fun). There is a post by Nil Mamano (used as inspiration) that describes a few approaches we can use during an interview to handle the case when we desperately need a queue, but we realize we're using JavaScript for a coding session. There are a few poorly implemented queues in JavaScript, as well as a few well-implemented ones. Google search lists the video from Monsterlessons Academy as one of the top results, and the explanation is quite nice. Check it to avoid problems with implementation. The post: https://lnkd.in/dk9s7AFV The problem: https://lnkd.in/d7FS-Ryq The Nil Mamano post: https://lnkd.in/duVitthA The Monsterlessons Academy video: https://lnkd.in/d5UmWeua #LeetCode #DSATips #DSA #ThinkDSA #ProblemSolving

1

Another day, another working theory which I thought was rubbish and turns out to maybe be a thing. This one started in grappling, where being able to maintain your ability to turn will stop you from getting submitted and leaked over into software, where your ability to rollback/change will stop you from taking big risks. It had a gap though - always changing means you never commit, right? Actually, no. It's about maintaining your ability to get to the next "knowable", without overreaching. In grappling, this is a safe position you can consolidate and progress. In software this is only changing what you need for the next feature. Thanks for listening, be sure to tip your waitress.

2

Elite R&D teams have this approach towards software engineering and estimations...it's dreamland and caters for multiple types of stakeholders if you have their buy-in🎖️

5

It's that time again! I've got a talk up this year that started out as a discussion in a DigiLocal CIO session about the best ways of talking to young people about testing the code you've written and led to me writing a basic testing framework in Scratch and seeing how feasible TDD was in that environment. The talk is called "Testing from Scratch" and if that sounds in any way interesting I'd really appreciate any votes to get it into the programme, but there are loads of great talks on offer (in particular, I'm curious to see what both Lorraine Pearce and Michael Luke are talking about) so do have a scroll and see all the wonderful things on offer

8

1 Comment

The wonderful Oliver Whitwam has written a short blog about how we're practicing type-driven development and domain-modelling to create software that is both correct and is communicable via ubiquitous language. We're really making good usage of the Rust newtype and Haskell-esque smart-constructor patterns to leverage types to encode business rules and invariants (until Rust adds refinement types via pattern-types!). If you love #types, #correctness, #domainmodelling, you'll like this blog I'm sure! https://lnkd.in/eqKWXtqS

17

1 Comment

👻 Kiro 👻 has had a whole host of great new features recently. There are the really big ones like Skills support and the constant evolution with new models like Opus and Sonnet 4.6, but one of my favourites is ACP... The Agent Client Protocol lets you take advantage of one of the best coding assistants 💪 around and use the underlying agent in a variety of different clients. I've been exploring how you can take advantage of Kiro for use cases that aren't all about writing code 💻 and how you and your teams can all reap the benefits from its huge capability. I've got my latest post in the comments on some of my favourite new ways to use Kiro with ACP. Let me know what you think. 🚀 #kiro #genai #kirocli

27

2 Comments

Managing developers isn’t about checking in every five minutes... It’s about giving them the right environment to do their best work. Here’s what that actually means: 🔷Be crystal clear on what needs doing - outcomes, not tasks 🔷Give proper context - the why behind the work 🔷Remove blockers - don’t be one 🔷Let them get on with it - deep work needs space 🔷Trust them to find better ways of doing things Developers aren’t looking for micromanagement. They want autonomy, purpose, and a team that knows how to deliver without chaos. If they’re constantly waiting on approvals, chasing clarity, or stuck in meetings... They won’t stick around for long. Lead with clarity. Support with trust. And manage based on results, not noise. Simple, but often missed we are djr 🔵 ⚪

7

5 Comments

Rare occasion of me posting something on LinkedIn full of experts and opinions, so I’m trying to share something genuinely useful. Over the last few months I’ve been very active in agentic engineering and kept running into tooling gaps I had to close for myself (and for a few people from the community). 1. Holding a local environment together with agentic tools is still annoying and doesn’t give enough control or flexibility. If I control execution myself, I waste time on terminal babysitting. If Cursor/Claude controls it, I lose visibility: I can’t properly read logs or understand what’s going on. So I built Crux, an AI-optimized local tool to run your app and keep it observable. The idea is simple: generate a config with an LLM, run crux in the terminal, and it starts your backend, frontend, mobile app, and dependencies, with logs you can actually follow. The main “magic” is built-in MCP support, so your agentic tool can also control the execution lifecycle: read logs, restart an instance, or trigger hot reload for a Flutter app. Tested on macOS only for now. If you make it work on Linux, contributions are welcome. https://lnkd.in/gnSAnWji [1/2]

6

Code agents could finally end the framework wars (looking at you, javascript). I've built a couple of web apps recently using claude code, both of them use vanilla js for the whole front end. No heavy framework required, keeping the website snappy and absolutely functional. It's so refreshing, and perfectly optimised for what I need. There are abstractions but they're where it makes sense for my app, not what a generic framework needs. I read a post recently that agents helping you code would inevitably lead to lower quality. I don't agree. I think being able to produce solid applications in languages like go and just vanilla js with plain css (no bootstrap!) will change things for the better.

21

7 Comments