Notes on building systems that survive contact with a real business
No trend pieces. Practical writing on automation, AI agents and integration decisions — the trade-offs we make on actual projects.
- AI2026-05-19
Retrieval that stays accurate after month one
Most RAG demos work. Most RAG systems rot. The difference is who owns the corpus and how updates reach the index.
8 min read - AI2026-05-05
What an AI feature actually costs to run each month
Model pricing is the small part. Retries, context bloat, evaluation and human review are what turn a cheap prototype into an expensive habit.
6 min read - Integration2026-04-28
Connecting an ERP without destabilising it
The ERP is the system of record and it is allowed to be slow and strict. Integrating with it well means respecting that, not fighting it.
7 min read - Ecommerce2026-04-07
The backend work behind an ecommerce conversion rate
Most conversion advice is about buttons. The larger losses usually sit in stock accuracy, delivery promises and payment failures.
7 min read - Data2026-03-31
Your AI project is mostly a data cleanup project
Models are commoditised. Whether your catalogue, CRM and documents are coherent is what decides the result.
6 min read - Strategy2026-03-10
Replacing a system nobody dares switch off
Big-bang rewrites fail for predictable reasons. Strangling a legacy system route by route is slower to plan and far safer to finish.
8 min read - Security2026-02-17
The security baseline every business system needs
You do not need an enterprise programme. You do need identity, least privilege, backups you have tested and an audit trail.
6 min read - Data2026-01-27
Operational reporting people actually open
Most dashboards are abandoned within a month. The ones that survive answer a decision somebody makes on a schedule.
5 min read - Delivery2026-01-13
Building Arabic-first products that do not feel translated
Mirroring a layout is the easy half. Numbers, typography, content length and search behaviour are where bilingual products usually fail.
7 min read - Ecommerce2025-12-16
Choosing payment methods for a market, not for a demo
Card coverage varies enormously by country. The right mix of methods is usually worth more than a checkout redesign.
6 min read - Strategy2025-11-25
How to evaluate a software partner before you commit
Portfolios are marketing. Ask about handover, ownership, response times and what happened on the project that went badly.
6 min read - AI2026-06-18
An AI agent is not a chatbot — and the difference costs money
Most disappointing AI projects fail on the same boundary: the thing was allowed to talk but never allowed to act, or given data it could not verify.
7 min read - Automation2026-05-27
How to tell which manual process is worth automating
Frequency times duration times error cost. The arithmetic is simple; the discipline is in refusing the processes that fail it.
6 min read - Integration2026-05-06
What actually goes into a serious WhatsApp integration
The link in the header is the easy part. The value is in templates, sessions, routing and keeping a searchable history.
6 min read - Strategy2026-04-14
Buy, configure, or build: an honest decision framework
Custom software is not automatically the answer, and neither is another subscription. The question is where your advantage actually lives.
7 min read - Integration2026-03-23
Why integrations break, and what a durable one looks like
Retries, idempotency and reconciliation are not optional extras. They are the difference between a demo and a system.
6 min read - Delivery2026-02-26
How we get a useful first release out in weeks
Not by working faster. By choosing a narrower first cut and refusing to build anything the first users will not touch.
5 min read
Tell us what you need to build.
Small integration, internal tool, AI feature or a full business platform — start with the problem, not a specification.
No specification needed — describe the problem in a few lines.