tecMAMBO CMS

Topic: AI

  • Gemini Spark review: promising, not quite ready

    Gemini Spark is one of the first personal AI agents built for ordinary people rather than developers, and it is genuinely useful for routine digital chores. But it asks for trust it has not fully earned yet, it costs premium money, and it still needs supervision. Promising, worth watching, not yet worth relying on.

    Google introduced Spark as a personal agent inside the Gemini app: a tool that does not just answer you but goes off and does things, building custom workflows and continuing to work in the cloud even when your phone is locked. After spending time with it, the short story is that the idea is right and the execution is early.

    ## What it does well

    The core promise lands more often than I expected. For routine, low-stakes tasks, organising and triaging, drafting and queuing things up, pulling together information from across your day, Spark can genuinely take work off your plate and keep going in the background. When it works, it feels less like using an app and more like having handed something to a capable assistant. That is a real shift, and at consumer scale it is new.

    ## Where it stumbles

    The trouble is the same trouble every agent has: it acts, and acting means it can act wrongly. A few times it confidently did the not-quite-right thing, which meant I could never fully stop watching, and an agent you have to supervise constantly is only half an agent. It is also early in obvious ways, with rough edges and behaviour that varies. And it sits behind Google’s premium subscription, so you are paying top prices to use something that still asks for your patience.

    ## Who it is for

    Spark is for the curious and the comfortable: people who enjoy being early, already pay for Google’s top AI tier, and have low-stakes tasks they are happy to delegate and double-check. If you want something dependable that you can set and forget, wait. This is a first chapter, and a promising one, but it is not the finished book.

  • Best AI subscription for your money in 2026

    The honest starting point for AI subscriptions in 2026 is this: for a lot of people, the free tiers are already enough. The big assistants give away a genuinely capable version, and unless you are using AI hard every day, you may be about to pay for power you will never touch. So before any recommendation, here is the question to answer: how often do you actually use this, and for what?

    Let us break the market into three tiers and match them to real people.

    ## Free: enough for most casual users

    The free versions of the major assistants now handle everyday writing, summarising, brainstorming, and quick questions well. Cost: nothing. If you reach for AI a few times a week to draft a message, tidy some text, or get unstuck, stay free. You are not missing much, and you are spending nothing.

    ## Cheap mid-tiers: for daily users on a budget

    A new layer has appeared at the low end, with at least one major assistant offering a paid tier at around USD 4.99 a month. For someone who uses AI most days but does not need the absolute top models, this is the sweet spot: more capacity and fewer limits, without the premium price. Best value pick for a daily user who wants more than free but does not want to spend like a professional.

    ## Premium: only if AI is core to your work

    The flagship plans, commonly around USD 20 a month and up, unlock the most capable models and the highest limits. These earn their keep only if AI is genuinely central to how you work, a writer, coder, analyst, or builder who would feel the difference every day. If that is you, the cost is easy to justify. If it is not, you are paying for a professional tool to do occasional chores.

    ## The money math people forget

    Two things matter especially in Kenya. First, most of these prices are in dollars, so the real cost in shillings moves with the exchange rate, and a cheap plan abroad is less cheap here. Second, these are cloud tools, so on mobile they also cost you data. Factor both in before you subscribe. A useful trick: add up what a year of any plan costs in shillings, then ask whether the tool clearly saves you more than that in time or money. If you cannot answer yes quickly, stay on the tier below.

    ## One more option for the technically inclined

    If you are comfortable with a bit of setup, open models such as Gemma and Qwen can be used at low or no cost, and they keep getting better. They will not always match the top paid models, but for many everyday tasks they are more than good enough, and they put a sensible ceiling on how much anyone should pay for the basics.

    ## The bottom line

    Best free: stick with a major assistant’s free tier if you use AI occasionally. Best value: a cheap mid-tier around five dollars a month if you use AI most days. Premium: only if AI is core to your work and you feel the difference daily. Buy the tier that matches how you actually use it, not the one the marketing says you need.

    Disclosure: tecMAMBO may earn a commission from some links, which never affects our recommendations. Prices and plans change often.

  • What is an AI agent, really?

    An AI agent is software that does not just answer you, it takes actions to reach a goal you set, with some independence along the way. That is the whole idea in one sentence. Everything else is detail.

    Here is an analogy that holds up well. A chatbot is like a knowledgeable friend you ask a question: you get a good answer, and then it is back to you to do something with it. An agent is more like an intern you hand a task to. You say what you want, and it goes off, makes a plan, uses the tools it has, checks its own work, and comes back when the job is done. The difference is not how clever the answer sounds. It is whether the thing acts.

    In 2026 you are meeting agents whether you sought them out or not. Google’s Gemini Spark is pitched as a personal agent that runs tasks in the background. Anthropic lets teams hand tasks to Claude and walk away while it works. Coding tools now run as agents that write, test, and fix code on their own. The pattern under all of them is the same.

    ## What actually makes something an agent

    Strip away the branding and a real agent usually has four things working together: a goal you give it, some autonomy to decide the steps, tools it can use, and a loop where it plans, acts, checks the result, and tries again if something went wrong.

    If a product can plan a multi-step task, use tools, and recover when a step fails, it is fair to call it an agent. If it just answers questions in a chat window, it is a chatbot, no matter what the launch slides say.

    ## Why the excitement, and why the caution

    The excitement is real. An agent that can quietly handle the boring, repetitive parts of your digital life is genuinely useful, and for a small team it can feel like extra hands. That is why every big company is racing to ship one.

    The caution is just as real, and it is the part the marketing skips. An agent acts, which means it can act wrongly, at speed, and at scale. To be useful it usually needs access to your accounts, your files, or your tools, and the more it can touch, the more a mistake can cost. The sensible posture in 2026 is to let agents handle low-stakes, reversible chores, and to keep a human hand on anything that spends money, sends messages on your behalf, or cannot be easily undone.

    ## The one-question test

    Next time something is sold to you as an AI agent, ask one thing: can it take a multi-step action on its own and recover when a step fails? If yes, it is an agent, and you should think about trust and access before you switch it on. If no, it is a chatbot with a new sticker, and you should not pay agent prices for it.

  • Kenya’s AI rules mean more than paperwork

    AI policy can sound distant until a startup tries to sell a tool to a bank, a hospital, or a county office. Then the questions get practical very quickly. Where is the data stored? Who is accountable when the answer is wrong? Can a person appeal a decision the system helped make?

    Kenya’s opportunity is to keep those questions practical. A rulebook that is too loose leaves citizens exposed and serious buyers nervous. A rulebook that is too heavy can make young companies spend more time proving compliance than proving usefulness.

    The best version sits in the middle: clear consent, clear accountability, room for local experimentation, and enough certainty that builders do not have to wait for rules written somewhere else.

  • Why AI hallucinates, and how to catch it

    An AI hallucination is when a chatbot states something false as if it were true, fluently and with total confidence. The unsettling part is not that it gets things wrong. It is that it gets them wrong in the same calm, polished voice it uses when it is right.

    To see why this happens, it helps to know what these tools actually do. A chatbot is not looking up answers in a database. It is predicting likely next words, one after another, based on patterns it learned from enormous amounts of text. Most of the time those patterns line up with reality, so the answer is correct. But when it hits a gap, a fact it does not have, a source that does not exist, a number it never saw, it does not stop and say so. It fills the gap with the most plausible-sounding words, because that is what it was built to do.

    ## Where it bites hardest

    A few situations produce hallucinations again and again: made-up sources, confident numbers, invented quotes, and hyper-local detail. Ask about a specific Nairobi street, a local law, or a small company, and the odds of a smooth, wrong answer go up, because the model has seen less reliable text about it.

    ## How to catch it before it catches you

    You do not need to be technical to stay safe. Ask for sources, then actually check them. Be most suspicious of exact figures, dates, names, and quotes. Cross-check anything you will rely on. Prefer tools that ground their answers in search and cite real pages. Treat AI as a fast first draft, not the final word.

    This, by the way, is why tecMAMBO does not let AI invent facts in our work, and why a person checks everything we publish. The same standard serves you well: let AI speed you up, but keep the judgement human.

  • Tokens and context windows, why AI forgets

    A token is a chunk of text an AI reads and writes, and a context window is how much of that text it can hold in mind at once. Get those two ideas and a lot of confusing chatbot behaviour suddenly makes sense.

    Start with tokens. AI does not read whole words the way we do. It breaks text into small pieces called tokens, which are often words or parts of words. A short message is a handful of tokens. A long document is thousands. Everything the model reads from you, and everything it writes back, is counted in tokens. That counting is also how most AI tools bill: you pay per token in, and per token out.

    Now the context window. This is the amount of text, measured in tokens, that the model can pay attention to at one time. Picture a desk. The context window is how much paper fits on it. While your conversation is small, everything sits on the desk and the model can see it all. As the chat grows, the desk fills up, and to make room, the oldest papers slide off the edge. That is the moment your chatbot forgets what you said at the start.

    In 2026 these desks have become enormous. Top models advertise context windows of a million tokens or more, enough to hold whole books. That is genuinely useful. But bigger is not automatically better. A huge window costs more to use, and stuffing it with everything can actually make a model less focused, the way a cluttered desk makes it harder to find the one page that matters.

    ## How to get better answers

    Put the important bit close to your question. Summarise long threads. Start fresh for a new topic. Do not over-stuff the prompt with the entire folder unless the task truly needs all of it.

  • Anthropic’s top AI models pulled over a US export order

    Anthropic has suspended access to its two most capable models, Claude Fable 5 and Claude Mythos 5, after the United States government issued an export-control directive. According to Anthropic’s own statement, complying with the order meant turning the two Mythos-class models off for customers while other models, including Opus, Sonnet, and Haiku, kept working normally.

    Here is the plain version of what these models were. In early June, Anthropic introduced a new top tier that sits above its established Opus line, with Fable 5 as the widely released version and Mythos 5 offered only to a small set of organisations. Days later, the export-control order arrived, and the most powerful options came off the table.

    If you use Claude through the app or through everyday tools, this most likely does not change your day. The models ordinary users reach are unaffected. So why pay attention?

    Because it tells you how governments now see frontier AI. The most capable models are being treated less like ordinary software and more like strategic technology, in the same bracket as advanced chips, where a single policy decision can switch off access overnight. That logic does not stop at one company or one country.

    For anyone building on AI in Kenya, there is a quiet lesson in here. If a tool can be switched off by a decision made far away, it is risky to wire your most important work to one top-tier model from one provider. The teams that cope best will be the ones that can swap models without rebuilding everything, and that keep a sensible fallback.

    This is a developing story, and the exact access status for the top models may continue to change. The bigger point is already clear: frontier AI is now part product, part policy question.

  • You can now tag an AI in Slack like a colleague

    Anthropic has launched Claude Tag, a way to bring its AI into Slack so a team can mention it in a channel and hand it tasks. You type @Claude, describe what you need, and it works in the background while you get on with something else. It can remember relevant context from the channels it sits in, and it can plan and carry out tasks over time rather than only answering in the moment.

    The shift here is subtle but real. Most people still use AI like a very fast search box: you ask, it answers, you move on. Claude Tag pushes a different habit, closer to handing a job to a colleague and trusting them to come back when it is done.

    Should you care? If you work in a small team, this is the more interesting half of the AI story. The promise is that a few people can take on work that used to need more hands, by delegating routine tasks to an assistant that runs while they sleep. For a lean Nairobi startup, that is the kind of leverage that actually matters.

    The caveats are worth stating plainly. Claude Tag is aimed at Team and Enterprise customers, not casual free users. And because you are granting an assistant access to channels and tools, who can see what, and what the assistant is allowed to touch, becomes a real decision, not an afterthought. Anthropic lets administrators scope that access tightly, which you should use.

    The takeaway is less about Slack and more about direction. AI is moving from a thing you talk to toward a thing you delegate to. This is an early, visible step in that move.

  • Google’s Gemini gets an agent, a video maker, and a brief

    At its I/O 2026 event, Google announced a wave of Gemini updates aimed at turning the app from a chatbot into an all-purpose assistant that can act for you. The headline additions: Gemini Spark, described as a personal agent that keeps working in the background; Gemini Omni, a video model that turns prompts and media into generated video; and Daily Brief, which pulls your inbox, calendar, and key tasks into one morning digest.

    Google also rebuilt the app’s look and changed how answers are shown: instead of a wall of text, the key point appears at the top, with detail below. Readers of tecMAMBO will find that familiar, because leading with the point is the whole idea behind plain-English tech.

    The reason this matters more than a typical product update is distribution. Gemini is becoming the default assistant across Android and Google’s apps. When an agent and a video generator are built into tools you already use, you do not have to adopt them. They simply show up.

    Two things are worth watching. First, Spark is an agent, meaning it takes actions, not just answers, so the questions of trust and oversight that come with any agent apply here too. Second, Google is expanding its content-labelling tools, including SynthID and Content Credentials, to flag AI-generated content across more places. In a year when telling real from generated is getting harder, that labelling may end up being the most useful announcement of the lot.

  • The quiet change ending all-you-can-eat AI

    Anthropic has changed how it charges for one kind of Claude usage: automated, programmatic work, the sort that powers coding agents and scripts rather than a person typing in a chat. The important distinction is between interactive use, where a human is actually using the tool, and headless or automated use, where software can keep calling the model on its own.

    The reason is simple arithmetic. A person using AI sends maybe dozens of prompts a day. An autonomous agent can fire off thousands, run tests, and call the model again and again, burning far more compute than a flat monthly fee was ever designed to cover. That is why all-you-can-eat AI subscriptions may not survive the agent era.

    Should you care, even if you are not a developer? Yes, because it is a preview of where AI pricing is heading for everyone. As AI shifts from a thing you type into toward agents that run jobs on your behalf, billing shifts with it: away from a tidy flat fee and toward something metered, like data bundles or electricity, where heavy use costs more.

    The practical advice is to know which kind of user you are. If you chat with AI a few times a day, flat plans still suit you fine. If you start handing tasks to agents that run on their own, watch the meter, because that is where the real cost now lives.