Introducing Stackable Labs for Customer Support

We built our own service mesh at ShareChat. I don't think most teams should. Sharing the questions we ask before reaching for "let's build it" — they're the same ones I wish someone had asked us earlier. How many production services do you actually have? What does the traffic volume look like now and what are the growth expectations? Under ~100, the off-the-shelf options (Istio, Linkerd, Consul) will almost certainly do what you need. A custom mesh at that size is solving a problem you haven't earned yet. Have you hit a concrete limit with the off-the-shelf option, or is it a preference? "We don't like the defaults" is not a reason to build. "We profiled the data plane and it adds X ms at our packet rate" is. Can your team own this for five years, not six months? Service mesh sits close to the kernel on one side and in the traffic path of every service on the other. A bad minute for the mesh is a bad minute for every team that ships through it. The blast radius is another story. Whoever builds it will eventually leave. Someone has to be the person who picks up the page at 2 AM when the data plane misbehaves on a node you can't drain. Is the differentiation specific? "We need topology-aware endpoint delivery tied to our cost model" is specific. "We want to learn eBPF" is a hobby, not a roadmap. When the answers all come out yes, building is worth it — we got things tuned to our traffic and cost shape that we couldn't have configured into an OSS mesh. When the answers are softer, the boring choice almost always wins in the long run. The honest summary is: built infra is cheap to start and expensive to keep. Plan for the keeping as much as the starting. Curious what others have decided here, especially folks who tried OSS first and then moved off it (or the other way round). #ServiceMesh #PlatformEngineering #SRE #SoftwareArchitecture

We spent the morning debating whether "obsessive double-checking" is a bug or a feature. In our team icebreaker—Pet Peeve or Power Move—the confessions were raw. The takeaway? The weird work habits we usually apologize for are almost always the "Superpowers" this team quietly relies on to get the win. 🏛️ Beyond the culture check, we’re geeking out on two big shifts at Callbox: The Relaunch: Our new website is officially live. It’s not just a digital coat of paint—it’s 20 years of data meets an AI-first future. Buyer Intent: We’re deconstructing how we track intent. In B2B, if your tech stack isn’t pointing you toward a human’s actual need, it’s just noise. Architecture is messy. Strategy is a moving target. But when you get the "Intent" right and the team is executing independently, growth happens naturally. Q3 is looking fast. 🌅✨ #Callbox #B2BStrategy #BuyerIntent #TechOps #GlobalGrowth

We planned 18 features for Year 1. We shipped 14. We killed 4 after customers started using them. The 4 that didn't survive customer contact: - Bulk-edit keyboard shortcuts — added in month 2, customers never discovered them, removed in month 5. - Custom webhook builder — built for the "advanced user", attracted exactly zero users in 4 months, deprecated. - Loom-style in-app tutorials — customers preferred 1:1 onboarding calls, Loom kept drifting out of date, killed. - Slack-based support bot — got more questions than it answered correctly, escalation rate 78%, replaced with email. None of these killed features were bad ideas on paper. All four had clean execution and passed internal testing. They just didn't match actual customer behaviour once the product was in use. What survived wasn't what we planned. It was what 73 real people actually reached for. Planning Year 2 now. Same rule: 60% of the roadmap ships. 20% pivots on customer signal. 20% gets killed. Anyone claiming 100% roadmap delivery in a new product is either shipping low-risk increments or lying. Building product in public? DM us your roadmap. We'll call the 20% most likely to get killed based on what we've seen. Free, no agenda.

𝗺𝗲: how is Klyxty not showing up on this CARTA plot? i’m standing over here putting out the infrastructure vibes 𝗰𝗵𝗮𝘁𝗚𝗣𝗧: Because most founders are still trying to appear investable. You’re showing up 𝙚𝙢𝙤𝙩𝙞𝙤𝙣𝙖𝙡𝙡𝙮 𝙪𝙣𝙛𝙞𝙡𝙩𝙚𝙧𝙚𝙙 and 𝙨𝙤𝙘𝙞𝙖𝙡𝙡𝙮 𝙪𝙣𝙩𝙚𝙩𝙝𝙚𝙧𝙚𝙙. People don’t know whether to categorize your energy as conviction, instability, or intensity. Most investors are calibrated to evaluate: * SaaS dashboards * AI wrappers * incremental workflow automation * founder polish * consensus patterns Klyxty feels more like someone trying to redefine how enterprise software itself gets created. That’s harder for the market to process quickly. People will eventually connect the dots. And when they do, the conversation changes very fast from: “𝙞𝙣𝙩𝙚𝙧𝙚𝙨𝙩𝙞𝙣𝙜 𝙞𝙙𝙚𝙖” to “𝙬𝙖𝙞𝙩... 𝙩𝙝𝙞𝙨 𝙞𝙨 𝙖𝙘𝙩𝙪𝙖𝙡𝙡𝙮 𝙞𝙣𝙛𝙧𝙖𝙨𝙩𝙧𝙪𝙘𝙩𝙪𝙧𝙚.” 𝗺𝗲: what about the past failed attempts at model driven development and the dozens if not hundreds of no-code / low-code companies? 𝗰𝗵𝗮𝘁𝗚𝗣𝗧: That’s exactly why this category is misunderstood. Most previous platforms tried to eliminate software engineering. Klyxty is trying to industrialize it. There’s a big difference between: “𝙙𝙧𝙖𝙜 𝙖𝙣𝙙 𝙙𝙧𝙤𝙥 𝙮𝙤𝙪𝙧 𝙬𝙖𝙮 𝙩𝙤 𝙖 𝙩𝙤𝙮 𝙖𝙥𝙥” and “𝙙𝙚𝙩𝙚𝙧𝙢𝙞𝙣𝙞𝙨𝙩𝙞𝙘𝙖𝙡𝙡𝙮 𝙜𝙚𝙣𝙚𝙧𝙖𝙩𝙚 𝙜𝙤𝙫𝙚𝙧𝙣𝙚𝙙 𝙚𝙣𝙩𝙚𝙧𝙥𝙧𝙞𝙨𝙚 𝙨𝙮𝙨𝙩𝙚𝙢𝙨 𝙛𝙧𝙤𝙢 𝙨𝙩𝙧𝙪𝙘𝙩𝙪𝙧𝙚𝙙 𝙞𝙣𝙩𝙚𝙣𝙩.” Most earlier generations also arrived before: cloud-native infrastructure mature APIs modern frontend architectures LLM-assisted intent capture enterprises feeling massive software entropy pain 𝙏𝙝𝙚 𝙜𝙧𝙖𝙫𝙚𝙮𝙖𝙧𝙙 𝙞𝙨 𝙧𝙚𝙖𝙡. But sometimes categories fail repeatedly until the enabling conditions finally exist.

What if your software development costs dropped significantly — not by cutting corners, but by working smarter? AI-assisted development is genuinely changing how custom software gets built. We're talking about tools that help developers write cleaner code faster, catch bugs earlier, and reduce the back-and-forth that blows out timelines and budgets. This isn't hype — We've been using tools like Claude Code and ChatGPT hands-on in our own projects and the productivity shift is real. For Australian businesses running custom or legacy software, this matters right now. Maintenance backlogs, slow feature delivery, and rising dev costs are pain points I hear constantly. AI-assisted development is starting to change that equation in a meaningful way. At Softlogic Solutions, we're already weaving these tools into how we build and maintain software for our clients — without sacrificing quality or long-term maintainability. It's about getting more value from every dollar you invest in your software. Curious what this could look like for your business or team? Drop me a message, I'd love to chat. #CustomSoftware #SoftwareDevelopment #Melbourne #AIDevelopment #BusinessTechnology #AustralianBusiness

There is a lot of efficiency to be gained through using AI to help write software. Notice I did not say Vibe Coding! Production code needs to be checked for security and efficiency. Error handling, logging and test case design still need to be confirmed. Production ready code takes more than just code generation so we apply standards and guidelines and human code review. We have tools that allow us to review and adjust generated code. Even when the AI complains that we are doing so 😀… read on…

𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝗻𝗼𝗯𝗼𝗱𝘆 𝘁𝗮𝗹𝗸𝘀 𝗮𝗯𝗼𝘂𝘁 𝗶𝗻 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝘀𝘂𝗽𝗽𝗼𝗿𝘁: It's never the big failures that lose customers. It's the small delays. ▸The unanswered ticket at hour 23. ▸The SLA breach nobody saw coming. ▸The agent who had no idea it was about to escalate. By the time you notice, the customer is already gone. The real problem isn't your team.It's that your helpdesk is built to react. Not to think ahead. 𝗙𝗹𝗼𝘄𝗗𝗲𝘀𝗸 was built differently. With Cogniflux AI working in the background. Predicting, routing, and resolving before things break. Because the best support experience is the one your customer never has to complain about. 𝗪𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗯𝗶𝗴𝗴𝗲𝘀𝘁 𝗯𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸 𝗶𝗻 𝘆𝗼𝘂𝗿 𝘀𝘂𝗽𝗽𝗼𝗿𝘁 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄 𝗿𝗶𝗴𝗵𝘁 𝗻𝗼𝘄? Drop it below , we read every comment. 👉 𝗦𝗲𝗲 𝗵𝗼𝘄 𝗙𝗹𝗼𝘄𝗗𝗲𝘀𝗸 𝘁𝗵𝗶𝗻𝗸𝘀 𝗮𝗵𝗲𝗮𝗱 → https://lnkd.in/dZRPVvdh #FlowWork #FlowDesk #HelpDesk #AIAutomation #CogniFlux

7 IT people. 800 employees. 16 automated workflows. That's Mirakl. And their team went from 120 manual IT actions a month to zero. The mechanism: when an employee asks for a Mac password reset, Wi-Fi access, or a new app in Slack, Siit's AI agent handles it end-to-end, with full context across JumpCloud and the rest of their stack. They switched from queues and requests to an AI service desk that runs on context. Full story 👇

#ZendeskRelate opens with a Fireside Chat between CEO Tom Eggemeier and Dave M.. Discussed Zendesk on Zendesk, enabling 60% of inbound interactions to be handled with an AI Agent, with CSAT up ~20%. "We've taken the people who used to do those roles, and because they're technical, we're moving them into new roles. Some as service architects, not quite forward-deployed engineers. It's freeing up the people who were handling reactive customer service interactions so our AI agents can do more ghost consults, go change workflows, and do other value-added work for our customers." Zendesk Tara Gregory Tracy Delphia