Ideas worth
shipping.
Writing on AI enablement, software delivery, and the work that actually moves the needle. No fluff — just what we've learned building with teams.
Artificial intelligence enablement: why the training course is the part that does not work
Companies name the skills gap as their biggest AI barrier and then respond with education, which is the most common answer and the least effective one. Enablement is not teaching people the tool. It is changing what the job consists of, and it has four parts, only one of which is training.
AI integration with legacy systems: what to do when there is no API
The system holding the data you need is fifteen years old, runs on a server in your building, and has no API worth the name. That is the normal case, not the awkward exception, and there are four ways in with very different price tags and risks attached.
AI system integration: the five things you are actually connecting
Most quotes for AI system integration price one connection: getting the model to read something. There are five, and the other four are where the schedule goes. The connector layer standardised in December 2025, which solved the first problem and none of the rest.
Employee resistance to AI is usually a rational response
When an AI rollout produces nothing, the internal explanation is usually that staff resisted it. The measured evidence says otherwise: people are adopting AI faster than their employers realise, and where they hold back they are responding to something the company never said out loud.
AI software development: what actually changed if you are the one paying for it
Around 90% of technology professionals now use AI in their work, and somebody has probably told you that custom software should therefore be much cheaper. The typing did get cheaper. The deciding, integrating, reviewing and maintaining did not, and the published data on what AI-assisted code does to a codebase over time is worth reading before you sign anything.
AI workflows: how to find the ones actually worth automating
Two-thirds of South African businesses are now using AI daily or experimenting with it. Far fewer have a single AI workflow running unattended in production. The gap is not tooling. It is that nobody sat down and worked out which jobs were worth automating, and the honest list is shorter than the sales deck suggests.
What is an AI agent, and does your business actually need one?
An AI agent is a system that decides its own next steps to reach a goal you set, rather than following steps you wrote. That difference sounds academic and is not: it changes the cost, the failure modes, and whether the thing is worth buying at all. Most of what is sold as an agent is not one.
The EU AI Act deadline moved. Here is what still lands on 2 August.
For two years the EU AI Act's big date was 2 August 2026. It no longer is, at least not for high-risk systems, which have moved to December 2027. The Article 50 transparency obligations have mostly not moved, and they allocate duties by role rather than to everyone who touches AI. The corrected timeline, and who actually owes what.
AI integration, explained: connecting AI to the systems you already run
AI integration is the work of connecting AI to the tools, data, and workflows you already run, so it does real work in your business rather than sitting in a separate app. Here is what it involves in practice.
What AI actually costs a small business
Answers to 'what does AI cost' range from 'free' to seven figures, and both are useless to a business doing R15 to R50 million in revenue trying to work out what AI consulting should actually cost. The shape of the cost has changed, and the thing worth being afraid of spending money on is not what you would expect.
How to choose an AI consulting firm: a buyer's checklist
Choosing an AI consulting firm comes down to independence, a production track record, and how seriously they take your data. Here is the checklist to run before you sign.
What AI consulting actually means for a business like yours
Everyone is talking about AI enablement as if the meaning were settled. For an owner running a real business with no data scientist, it mostly isn't. Here is what AI consulting actually means, in plain language, minus the hype.
What does AI consulting cost in South Africa?
What AI consulting costs in South Africa depends far more on the engagement model than on the technology. Here are the four common models, with market rand ranges (ex-VAT), so you can budget honestly.
Your data moat is not your source of truth
'Data moat' is one of the most repeated phrases in AI strategy and one of the least examined. A moat is real and worth building, but it is not where your truth lives. The systems your data came from are. Here is what a data moat actually is, what it is not, and how to use one without quietly breaking your own architecture.
What does an AI consultant actually do?
An AI consultant helps you decide where AI genuinely fits your business, then gets it working in production. Here is what good AI consulting looks like, and how to spot the demo-theatre version.
Start from the decision, not the data
Most AI projects start from the wrong end: 'we have data' or 'we should use AI', and produce insight that nobody acts on. The teams that get returns start from a specific decision a human needs to make, and work backwards to the technology.
The AI projects that quietly pay for themselves
When an owner finally decides to try AI, the instinct is to reach for the most visible thing: a chatbot customers talk to. It is usually the worst place to start. The projects that pay off early are the boring internal ones nobody posts about.
Your AI feature works. Nobody's using it.
Building an AI feature that works and getting people to use it are two different achievements, and the second is where most of the value quietly leaks away. The published gap between access and use is now around 60 percentage points, and it is not a technology problem.
Most AI advice is written for companies that don't look like yours
The published AI playbook assumes a data team, a research function, and an eight-figure budget. Almost none of it survives contact with a company doing R15 to R50 million that knows its trade cold and has no data scientist. Here is what actually changes at that scale.
Everyone in your company is looking at different numbers
Before AI can answer anything useful about your business, your systems have to agree on what a customer, an invoice, and 'revenue' actually are. In most mid-sized companies, they don't. That disagreement is the real bottleneck, not the model.
Why your AI should never be the source of truth
Hallucination is not a bug you tune away. It is a structural property of how language models work. The teams who trust AI in production are the ones who decided, deliberately, which parts of the system the model is never allowed to decide.
What working AI teams actually look like in 2026
Most published AI team structures describe how it should be done. This post is about what it actually looks like in the teams that are shipping: the roles, the ratios, the rhythms, and the small operational habits that separate working from theatre.
Measuring AI ROI without lying to yourself
Ninety-five percent of enterprise AI investment produced no measurable return last year. Most of the failure is not in the technology. It is in measurement: what teams chose to count, what they chose to ignore, and what they hoped nobody would ask about.
Buy the boring, build the unique: an AI infrastructure framework
Most teams building AI features in 2026 are building too much of their own infrastructure. Here is a practical framework for what should live in-house and what should not, and what the total cost actually looks like when you count honestly.
Your data is your moat, and most companies' data isn't ready for AI
The model you choose is not your differentiator. Your data is. And the published numbers on how few companies have data that AI can actually use are bracing: between five and seven percent, depending on whose research you read.
The case for boring AI infrastructure
The teams shipping reliable AI are not winning on model choice. They are winning on evals, observability, cost control, and the fallback paths that fire when the interesting parts break.
POPIA, GDPR, and AI: what South African product teams need to know in 2026
South African teams shipping AI features cannot ignore either POPIA at home or the EU AI Act when serving European customers. Here is the practical compliance picture as of mid-2026.
Agents vs. workflows: choosing the right pattern in 2026
Most things sold as agents are workflows wearing a costume. Knowing the difference matters because the trade-offs (cost, latency, debuggability, reliability) go in opposite directions.
Why your RAG demo doesn't survive contact with real users
A working demo with five curated documents is not a working system. Here is what breaks when real users, real documents, and real load arrive, and how to build for it from the start.
Working through something similar? Let's talk.
Most of what we write here comes from real engagements. If it resonates, a conversation costs nothing.
- Format 60-min call
- Output Written summary
- Commitment None required