Methodology, industry takes, and case studies. Three new posts a week. Mostly Janaka. Sometimes the team.
What AI actually changes about every stage of software development — and why the phase everyone ignores is now the most important one.
$ read the postAI for real estate in Sri Lanka: qualifying portal and WhatsApp enquiries, viewings and follow-up, from US$1,225 a month, every draft approved.
Questions to ask an AI vendor before you sign: 20 of them, in 5 groups, with the answers that should end the meeting and a 12-month total in one shape.
AI for logistics in Sri Lanka: status replies, document packs and chasing, from US$1,225 a month, with the customs entry left to a licensed agent.
How to run an AI pilot in 30 days: one workflow, one named approver, one number measured weekly, and the day 30 decision written before you start.
AI for law firms in Sri Lanka: the four tasks that go first, the line no agent crosses, and the sum against LKR 402,700 a month for one agent.
The benefits of AI in audit in Sri Lanka: four tasks an agent takes, the ISA 500 and 530 lines it cannot cross, and a season priced at US$1,225 a month.
An AI product management consultant decides what to build and what to hand to agents. Free first call, US$2,300 mapping sprint, agents at US$1,225 a month.
AI automation for hotels in Sri Lanka: same-hour enquiry replies in three languages, confirmations and review replies, from US$1,225 a month with approval.
Sri Lanka's Personal Data Protection Act, No. 9 of 2022 covers AI agents that read customer data: the duties, the risks, and the LKR 10 million penalty.
How to write a spec for AI agents: 5 parts, outcome, constraints, examples, acceptance checks and must-nots, with a worked example and our template.
AI agents vs chatbots: a chatbot answers, an agent updates the booking or raises the invoice. Gartner found only 14% of issues resolve in self-service.
Spec-first development in plain terms: a written spec, agents that build to it, review of every pull request. 84% of developers now use AI tools.
AI courses in Sri Lanka priced: LKR 45,100 for a 40-hour certificate up to LKR 1,425,000 for a two-year MSc, four course types, and what each one skips.
Human in the loop for AI agents: an AI agent approval rate is the share of drafts a named reviewer approves untouched, scored weekly over 30 days.
AI for retail in Sri Lanka: order confirmations, stock and price answers, and web shop updates, all behind an approver, at US$1,225 a month, LKR 402,700.
WhatsApp automation in Sri Lanka: the three flows customers accept, the two that get you blocked, and the cost against an agent at LKR 402,700 a month.
AI for clinics in Sri Lanka: the four front-desk tasks that go first, what never moves, and the sum against LKR 402,700 a month for one agent.
AI for small business in Sri Lanka: the five workflows that pay first, the break-even sum at a loaded wage of LKR 409 an hour, and which one to start with.
Gartner’s five-stage AI maturity model is usually drawn as an even climb. It isn’t. Four of the steps are gradual; one — Emerging to Operational — is a cliff, and it’s the only one that decides whether the AI budget ever comes back.
AI agency vs AI consultancy vs freelancer, decided by budget: mapping at US$2,300, agents at US$1,225 a month, freelancers at US$15 to US$40 an hour.
Buying Copilot licences transfers capability without transferring instruction. Two government trials, MIT, Gartner and BCG all circle the same missing artefact — the brief nobody ever wrote down.
AI engineer salary in Sri Lanka in 2026: senior roles advertise at LKR 400,000 to 600,000 a month, and the real cost is about 1.4 times that. Five routes.
You did everything 2022 told you to do — funnel, CRM, trust badges, SEO audit, maybe a Webflow rebuild. Inquiries are still flat. You didn't execute badly. You executed a playbook written for a reader who is leaving the building.
Why your best engineer became the bottleneck — and how to spot the same freeze in yourself.
Will AI replace accountants in Sri Lanka? No. But 69% of data processing time is automatable. The 12 practice tasks that go first, and what it costs.
It was a symptom of under-resourced teams pretending breadth was excellence. AI did not create unicorns. It revealed that most of what we called unicorn skills was translation work that never needed a human in the first place.
What the first real data on agentic software development tells us — and what it means for how teams should be working in 2026. 1. The most quoted law in software is…
AI companies in Sri Lanka: Clutch lists 25 in its AI category. Here are the four types, a five-question shortlist method, and what each one charges.
AI automation cost in Sri Lanka in 2026: studio rates from under US$25 an hour, projects from US$10,000, AI agents at US$1,225 a month, in LKR too.
The traditional PM role isn’t just changing — it’s being structurally replaced from the inside. One person, the right spec, and the right toolchain now does what five used to. Here’s what that actually looks like.
The MVP was always a workaround for expensive software. Vibe coding eliminated that excuse. Here’s why the prototype is now the spec — and why spec-first is the only rational model for what comes next.
Landing page, knowledge base, Chrome extension concept, and a web app funnel — all from a single tax computation spec.
A roster, attendance, and biometric platform for operations managers — spec’d from a spreadsheet problem, not a feature list.
An ethical reputation management platform for Dutch hospitality — built on a 12-hour head start, not a filter.
The conversation about AI-native software development is having the wrong argument.
I raised $100,000, burned $90,000 on development, and shipped to almost nobody. Here’s the ratio I wish I’d known — and why the economics of 2026 have made getting this wrong significantly more expensive.
A mid-market HVAC company went from whiteboard sketches to a working CRM + dispatch + field app in three weeks — without a single requirements workshop.
14 screens. 295 staff. Zero ambiguity. How a specification document became an interactive enterprise prototype.
SAFe felt like enterprise overhead for a 15-person studio. Then we added AI agents to the workflow — and discovered that the coordination layer SAFe was solving for doesn’t disappear when the team is small. It explodes.
AI didn’t create spec-driven development. It just made ignoring it unaffordable.
How structured artifacts living in Git are becoming the operating layer of AI-native software — and why there’s no single right way to do it yet. Field notes from a studio that rebuilt how it works from the ground up.
No posts in this category yet. New ones land every Mon, Wed, Fri.
We don't run a newsletter. New posts land in the RSS feed the moment they ship — Monday, Wednesday, Friday. Pipe it into whatever reader you actually open.