I should start by saying that Kagi is a paid product. For those who want to look at what other search engines are out there, Startpage and DuckDuckGo both offer great services free of charge.

Why pay for search?

As the oft-repeated adage goes: "if you're not paying for the product, you are the product". We all know that there's a reason Google can provide the services it does for free. We also all know that this is a bit of an oversimplification, but it's probably a good heuristic to apply to the landscape of software tools nonetheless.

I like the fact that when I search on Kagi, the first hit is actually a hit, not a sponsored result or ad. It feels as though I'm using a direct tool without any advertiser incentives muddying the waters.

In itself, that's not enough to justify the cost for me, but the bundling of the below features pushes me over the fence.

I've been using it as a daily driver for months and I've never once felt the need to switch back. Not one single withdrawal itch.

Lenses

Lenses focus search results down across a narrow field. For example, you can narrow your search to only include Academic or Forum results. Furthermore, you can create your own custom lenses. It's not a feature I tend to use very often, but a neat little one nonetheless.

As an example, I've created a Bearblog lens which only returns *.bearblog.dev sites:

What's really cool is that these lenses can be applied to the lookups undertaken by AI features, which neatly leads me onto the topic of...

AI

Kagi has Quick Answer which serves the same purpose as Google's AI Overview. I'm not going to pretend that it's quite as good or as fast, but it's more than good enough for me. The main thing I like about it is that it's toggle-able. It's off by default and if I want to use it, I can just affix a ? to the end of my query. It's a feature, not a new default mode of search forced onto users.

Kagi also has Assistant - an inference provider for a whole host of well known LLM models. I like the idea of subscribing to this service, instead of directly subscribing to a specific model provider (e.g. via OpenAI's app, or Kimi's app). It means I'm not locked in; I can experiment with whichever model I like best. I can save this preference in a Custom Assistant (CA) which allows me to provide a system prompt describing the behaviour for all interactions used when that specific CA is selected. This CA can then be given a custom name and neatly accessed via the search bar. For example, I've got a CA called jair (James + AI + Reasoning; very creative, I know) which looks like:

With a system prompt of:

Include references wherever possible. 
Express uncertainty wherever possible; you should express doubt and avoid over-confident, authoritative statements. 
Keep the formatting light; the quality of the information is more important than its presentation. 
Take on the role of a professional correspondent. Statements should be direct and to the point. Discard any pleasantries or unnecessary verbosity.  
Do not cater to my feelings whatsoever. 
If you lack the information required to give a comprehensive and correct answer, prompt me for clarification. 

So now I can go and search !jair what is the circumference of the moon and how do we know and it'll give me a result as per my settings above.

As well as flexibility, Kagi as an inference provider also allows for a higher degree of privacy. Not necessarily full privacy, but better than that of most alternatives.

The privacy policy for Moonshot's Kimi app (the model I use via Kagi) states:

1. Personal Information We Collect:

User Content: This includes prompts, audio, images, videos, files, and any content you input or generate while using our products and services. We process this information to provide and improve the Services, including training and optimizing our models. The legal basis for this processing may be our legitimate interests or your consent, depending on your jurisdiction.

Device and Usage Information: We collect information about your device and how you interact with the Services, such as:Device type, model, and operating system;Browser version and user agent;Unique device identifiers (such as device ID, MAC address);Conversation IDs and session identifiers;Network and telecommunications provider;Clipboard data (if applicable and permitted by your settings);Date and time of access, pages viewed, and interaction patterns.This information helps us monitor service performance, troubleshoot issues, and optimize user experience.

This is pretty cookie-cutter; OpenAI, Alibaba, Google, Anthropic, etc. all store user prompts. If they offer a feature to not do so, then it's generally not available within consumer-grade plans.

If every prompt is being recorded, then I can't ask certain things. For 95% of the time throughout daily life, I'm prompting something innocuous. But for that 5% when my prompt is personal or involves information that I'm not comfortable or willing to share, then I'm out of luck. Unless I run a model locally (which is a faff and not all that practical a lot of the time).

Also, note that they state that they can potentially collect clipboard information. As someone who uses a password manager and so relies on the clipboard to sign in to services from time to time, I find that a little scary.

Kagi uses APIs from various services, all of which are set up to have temporary data retention or zero data retention, depending on the model being used. Using Kimi via Kagi, I know that my prompts and associated context files aren't being stored, and that my clipboard isn't being read.

It's worth noting however that Kagi is just acting as the middleman, and so are subject to any policy changes affecting the APIs they use. Furthermore, they state in their privacy policy that prompts "may be retained for a short period of time as a part of request debugging".

It's also worth noting that other LLM providers to offer paid tiers which feature data retention controls.

Small Web

Kagi's small web, as per their blog post:

...typically refers to the non-commercial part of the web, crafted by individuals to express themselves or share knowledge without seeking any financial gain.

This is more of a gimmick than a competitive search feature, but I'd argue that it acts as a good litmus test for their philosophy of the company. That of user control, and an individually expressive, human-centric internet built by people - not by big tech encapsulating their users in their proprietary walled gardens.

To Conclude

I like Kagi. I recommend you give it a go if you haven't (they have a trial). If you find that it's not worth the cost paying for a plan, then by all means give Startpage a go; it used to be my search engine of choice.

Kagi does also have other features that I've not talked about: Translate, News, Summarizer, their own web browser, and a host of other quality-of-life features alongside their search service.

If you have any feedback or disagree with anything I've said, let me know!

A lot of the above experiences are anecdotal; they derive from personal experience. I've not done any systematic tests or comparisons between Kagi and Google Search.

‘If you don’t have clean water, you really don’t have anything.’

The post Contamination, climate change and political drama stall clean water for Colorado’s Arkansas Valley appeared first on High Country News.

Steve Yegge’s article about programmer burnout (“The AI Vampire”) along with Margaret Storey’s article about Cognitive Debt started an ongoing conversation about programmer fatigue and software quality—two topics that should be linked, but often aren’t. Steve argues that programming constantly with the help of agentic AI leds to burnout; it’s fast, it’s fun, but keeping up with your agents causes mental strain. He recommends programming with agents no more than 4 or 5 hours per day. I could cynically say that most software developers spend at most 20% of their time writing code, which leaves about an hour and a half for wrestling with agents—but that’s beside the point. Yegge’s point about burnout is important, and is in line with what friends have told me. At some point, you have to put the laptop down.

Storey makes a different point. Agentic engineering is great at creating software that works, but that you don’t quite understand. Like humans, agents can generate a lot of spaghetti code. They can “design” convoluted and inappropriate software structures—I hesitate to call them “architectures”; they’re what happens in the absence of architecture. Agents are very capable of creating technical debt—and not the kind of meaningful technical debt that lets you release a product on time with the knowledge that you need to make pay it back with interest. If nobody is looking hard at the code, the debt can grow without bounds, sort of like not checking your credit card balance. What’s worse—and this is Storey’s contribution—while that technical debt is growing, developers are losing track of the design, the structure, the architecture. She calls that “cognitive debt.” You don’t just have problems in the code; those problems are harder to find and fix than they should be because you’re unclear on the structure of the code you’re working with.

Other voices have made similar points. The Sonarsource blog writes about how AI is reshaping technical debt and creating new burdens, new kinds of toil. In “The Mythical Agent Month,” Wes McKinney links the problem of burnout to the introduction of “accidental complexity” and “agent scope creep,” while Tim O’Brien writes that while scope creep isn’t new, AI supersized its growth. And Addy Osmani writes about finding your parallel agent limit, coming to grips with what you’re capable of accomplishing without compromising your work or your life.

Cognitive debt and burnout aren’t new, alas. With or without AI, we’ve all stayed up to 4AM working on a bug that won’t go away or pursuing an interesting idea to its end. Sometimes that’s heroic, but AI threatens to turn it into a lifestyle. AI fatigue is real, as Siddhant Khare writes, and it’s something we need to talk about. When fatigued, it’s tempting to say “this works, it looks good, and it passes our tests” without considering how the code fits into the overall plan. With 10x code generation, you also get 10x the debt load, and that’s being optimistic. When the debt curve goes exponential, strategies for managing that debt are stressed past the breaking point.

The problem with cognitive debt is that it eventually makes new features and bug fixes difficult or impossible. The code has become so convoluted that it can’t be changed. I’ve certainly done that with hand-written code: added a feature without thinking enough about how the new code fit in, added some more code later, and then—when I needed to add a third feature—discovered that I’d created a problem that wouldn’t be simple to fix. The right stuff was there, but in the wrong places because I wasn’t thinking about the overall structure.

That’s a common enough problem with handwritten code; it’s almost always a problem with legacy code where the original developers and maintainers are no longer around. We need to realize that it’s also a problem with AI-generated code, which has been characterized as legacy code from the day it’s written. Somebody or something has to pay down the debt. As Storey writes, “velocity without understanding is not sustainable”: not for humans, not for machines. If you understand the structure of what you’re building, you can steer the AI away from creating a problem in the first place, or you can use it to author a fix. If you don’t understand the structure or can’t describe it to the AI, you’re lost.

Cognitive debt accumulates much more quickly when you’re burned out. Burnout has always been a problem for programmers, especially for those who really love programming: you stay up all night to solve a problem. And, while some programmers resist using AI to write code, those who use AI frequently find that it exacts the same toll: it’s hard to stop. It is its own kind of toil: toil that gives you a sense of accomplishment and fulfillment, but still leaves you empty.

Agents may not be subject to burnout, but the humans who control them are. Agents are quickly becoming more capable, but they still can’t maintain a sense of the shape and structure of a project over the long term. That’s our job. They can pay down technical debt, but only if properly guided; that’s also our job. And we won’t be able to do either if we’re burned out.

Violent crime fell sharply across the largest U.S. cities in early 2026, extending a nationwide decline that began after the pandemic-era crime spike.

Why it matters: Data from 67 major U.S. law enforcement agencies show violent crime fell across major categories during the first quarter compared with the same period in 2025.

• The declines show up across every major region, suggesting a systemic, nationwide trend.
• The quarterly reports collected by the Major Cities Chiefs Association have been a good measure of trends that are reflected in the annual FBI crime data released in the fall.

By the numbers: Homicides dropped 17.7%.

• Robberies fell 20.4%.
• Rapes declined 7.2%.
• Aggravated assaults decreased 4.8%.

Zoom in: Some of the nation's biggest cities posted especially dramatic homicide declines in the first three months of 2026.

• Among those that saw sizable percentage drops in homicide were Washington, D.C. (64.7%), Philadelphia (54%), San Diego (50%) and Memphis (34.4%).
• New York City experienced a 31.7% drop in homicides during Mayor Zohran Mamdani's first months in office.
• Los Angeles (23%) and Houston (36.4%) also posted homicide declines during the same period.

Between the lines: The new numbers complicate the political narrative around crime heading into the 2026 midterms. President Trump has repeatedly described major Democratic-led cities as gripped by violent crime.

• Data show many urban areas have become significantly safer over the last two years, with drops beginning in the second half of the Biden presidency and continuing under Trump.
• Trump cited violent crime as his reason for sending federal troops last year to Chicago, Portland, Washington, D.C., Memphis, and cities in California.

The intrigue: Aurora, Colorado — a city Trump repeatedly and falsely singled out as being overrun by Venezuelan immigrant gangs during the 2024 election — saw a 66.7% drop in homicides.

• In response to early reports that crime was dropping to record lows, the Trump administration has changed its tone and has begun touting the declines while crediting its policies.

Yes, but: The recovery remains uneven.

• Some cities still reported increases in certain violent crime categories even as overall violence fell.
• Minneapolis, Atlanta, and Virginia Beach, Va., were among the cities that posted overall increases in violent crime totals during the quarter, according to Axios' analysis of the MCCA data.

Police leaders also caution that crime trends can shift quickly heading into the summer months, when violence historically rises.

• Denver officials recently warned about a potential seasonal spike after a string of killings despite the city's broader downward trend, Axios Denver's Esteban L. Hernandez reports.

The bottom line: America's largest cities are continuing to get safer in 2026, even as crime remains one of the country's most politically potent issues.