Skip to content
AI Radar by Wagner Solution
CS
Menu

A practical perspective for individuals

Less monitoring. More value.

We turn new tools and features into one question: what can you do with them today, without unnecessary hype?

  • AvailabilityIs it actually out?
  • BenefitWhat work could this improve?
  • First stepWhat is worth verifying?
459 practical events in the selection 72 active sources Updated

Practical radar

What deserves your attention

Browse the archive →
worth noting Productivity

Spotify launches Taste Profile feature in the US for adjusting recommendations via natural language

Starting Wednesday, Spotify is launching the Taste Profile feature in the US for Premium subscribers (18+), allowing users to adjust the recommendation algorithm for music, podcasts, and audiobooks using natural language. The feature also affects Discover Weekly and Spotify Wrapped and is currently in beta.

What you get out of it

Spotify Premium users in the US can now use text to directly adjust how the algorithm recommends music, podcasts, and audiobooks, including removing unwanted patterns (e.g., music played while sleeping or listening by someone else on a shared account).

1 source
Details and sources →
worth noting Security

Adoption of AI scribe in Australian healthcare: concerns about errors in records and protection of patient data

According to the report by Digital Rights Watch, over 40 % of Australian doctors use AI scribe for consultation notes; up to 20 % of records contain errors affecting diagnosis, and there is a lack of transparency about how patient data is handled.

What you get out of it

According to the report, a patient whose consultation is recorded by an AI scribe tool has no assurance that the notes reflect what actually happened (errors, hallucinations, bias based on ethnicity or language), nor transparency about where their health data goes.

1 source
Details and sources →
Anthropic worth noting update New models

OpenAI claims that GPT-5.6 Sol outperformed Opus 5 on ARC-AGI-3 – but only with its own testing environment

OpenAI states that GPT-5.6 Sol achieved a score of 38.3 % on ARC-AGI-3 compared with 30.2 % for Opus 5 from Anthropic. However, this holds only with the custom API configuration used by OpenAI; in the official test, GPT-5.6 Sol scores only 7.8 %.

New: OpenAI achieves 38.3 % on ARC-AGI-3 with its own API setup (Retained Reasoning + Compaction); In the official harness without these features, GPT-5.6 Sol scores only 7.8 %; ARC-AGI-3 was deliberately designed without external aids to measure raw model performance; Retained Reasoning and Compaction are not part of the official testing environment; Benchmark results reflect both the model and the technical setup around it

What you get out of it

When reading benchmark rankings, you need to verify that the figures being compared come from the same testing environment – a difference of 7.8 % versus 38.3 % for the same model shows that a score alone, without methodological context, can be misleading.

1 source
Details and sources →
worth noting AI agents

Simon Willison: the key skill with coding agents is giving instructions and verifying results, not reading every line

Simon Willison argues that a key skill when working with coding agents is being able to give them clear instructions and then verify the correctness of the changes made – and that reading every line of code is not the most effective way to do this.

What you get out of it

Developers working with coding agents should develop two specific skills – giving the agent clear instructions and verifying that the changes made match the task – while recognizing that reading every line of code is neither the only nor the most effective way to verify this.

1 source
Details and sources →
worth noting Productivity

Delegating work to generative AI carries a hidden risk of fabricated facts and declining critical thinking

The cases involving the South African government and Deloitte showed that AI can invent citations and authors. A survey of 48 000 people found that 66 % of them use AI regularly, but 61 % have no training and the main motivation is fear of falling behind, not quality.

What you get out of it

Anyone who uses AI outputs without verifying the underlying source material will not recognize the risk of fabricated facts or citations until it causes a problem – and without training (which 61 % of users lack, according to the cited survey), these blind spots are harder to recognize.

1 source
Details and sources →
context Other

A typology of four kinds of answers from AI systems for assessing their reliability

A librarian at University of Virginia proposes dividing answers from AI systems into four types – factual, interpretive, constructive and strategic – with a different verification process for each.

What you get out of it

When making a decision based on an answer from an AI system (e.g. to a health or personal question), first identify the type of answer and then choose how to verify it, rather than relying on its fluent and persuasive tone.

1 source
Details and sources →
context Security

Al Gore: the main risk of AI is the automation of work, not data center emissions

Al Gore said in an interview with TechCrunch that emissions from AI data centers are negligible compared with those from air conditioning and landfills; in his view, the real threat lies in warnings from leaders at OpenAI and Anthropic about job losses due to automation.

What you get out of it

Increasing warnings from experts at OpenAI and Anthropic about the impact of automation on jobs, which Gore describes as sincere, suggest that the expert community is taking the risk of job losses due to AI increasingly seriously, rather than viewing it merely as marketing rhetoric.

1 source
Details and sources →
OpenAI context update Coding

OpenAI advises simplifying skills and guardrails for the GPT-6 Astra model

Eric Provencher from OpenAI issued recommendations on 12. 9. 2026 on how to shorten skill descriptions, limit mandatory documentation reading in AGENTS.md and reconsider guardrails for the GPT-6 Astra model—according to the company, accumulated rules unnecessarily take up context and stop the model's work prematurely.

New: Eric Provencher from OpenAI issued specific recommendations for simplifying prompts and guardrails; Shorter and more specific prompts improve the performance of the GPT-6 Astra model; Unnecessary guardrails and blanket rules hamper performance; Skills are Markdown files containing instructions that are selected based on their descriptions; Long skill descriptions lead to poor selections and context for you

What you get out of it

Developers working with Codex and the GPT-6 Astra model should shorten and clarify their custom skill descriptions during the transition and check whether approval rules unnecessarily restrict the model.

1 source
Details and sources →
Meta context Security

Meta ran ads for the Kromix app, which generates non-consensual pornographic deepfakes of female politicians

Meta ran ads for the Kromix app, which generates pornographic deepfake videos without consent, depicting women including an American female politician. The ads violate the company's own rules against sexual content.

What you get out of it

Tools such as Kromix show that even public figures, including female politicians, can be depicted without their consent in pornographic deepfake videos distributed through paid ads on a major platform.

1 source
Details and sources →
Anthropic context Other

Claude model API outage lasting over two hours is resolved

On 18 August 2026, Anthropic addressed a Claude API outage involving increased error rates and degraded performance across multiple Claude models; the impact lasted 16:11–18:23 UTC and was resolved at 19:01 UTC.

What you get out of it

Anyone using Claude API during the period 16:11–18:23 UTC on 18 August 2026 may have experienced request errors or slower responses; the problem is now resolved and no further action is necessary.

✓ official
Details and sources →
context AI agents

Simon Willison: coding agents make software engineering harder, not easier

Independent commentator Simon Willison claims that the more he works with coding agents, the more convinced he is that they make software engineering harder. According to him, their full potential is unlocked only by exceptional discipline and knowledge.

What you get out of it

A developer using coding agents should expect that their benefit does not come automatically, but requires sustained discipline in checking and managing their outputs.

1 source
Details and sources →
context Tools and apps

AI photo editors are moving from editing to reconstructing image content

According to Root.cz, AI photo editors are ceasing to be passive tools and beginning to actively reconstruct image content – computing details and smoothing textures without metadata distinguishing this from the original capture.

What you get out of it

Photographers should be aware that common AI features in editors (filling in details, smoothing textures) are no longer just technical corrections, but may actually compute or create part of an image without the resulting file or metadata clearly distinguishing it from the original capture.

1 source
Details and sources →

When personal experience proves useful

Does it make sense for a team too?

Look at the same topics through the lens of processes, risks, and implementation within a company.

Go to the company radar →