Artem Techman retweeted
You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own.
Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!
Artem Techman retweeted
500+ Claude skills. $99 a year. Full Access.
Handpicked from Anthropic, trusted vendors and creators. Every one scanned with NVIDIA SkillSpector for prompt injection, data exfiltration and destructive code before listing.
Free for 2 days, then $99/year.
skillsaggregator.com
🤖 Made with AI
How much does a corporate AI assistant really cost?
Imagine a company wants to deploy an AI assistant for 1,000 employees.
When discussing the budget, people often start with the cost of the model: how much the tokens, API, or GPU server cost.
But that is only a small part of all the expenses 🤩
Here is what the real cost consists of:
Model and infrastructure
API or your own environment, compute resources, data storage, redundancy, and load monitoring.
Corporate data
Document preparation, search organization, indexing, source updates, and access rights management.
Integrations
Connecting CRM, ERP, the corporate portal, document management systems, and other data sources.
Security
Data transfer controls, operation logging, access management, and verification of AI responses and actions.
Operations
Support, quality testing, model updates, incident handling, and user training.
How to assess the economics? 🤔(MOST IMPORTANT PART)
Let’s take a hypothetical scenario.
1,000 employees use the AI assistant, and each saves an average of 15 minutes per working day.
Over 20 working days, the potential savings amount to:
5,000 working hours per month.
At a conditional hourly cost of 15 EUR, this equals 75,000 EUR in labor costs per month.
But this is not yet monetary savings.
To turn the freed-up time into a financial result, you need to understand which operations are actually reduced and how the company uses the released capacity.
The main mistake is comparing the cost of AI with potentially saved hours without verifying the real effect.
For CTOs and CIOs, the right question sounds different:
How much does the entire lifecycle of the AI solution cost, and what confirmed value does it create?
#corporateai #crm #erm #enterprise #enterpriseAI #anthropic #mcp #api #bedrock #amazon #azure
🤖 Made with AI
2
53
Artem Techman retweeted
BREAKING: World's largest crypto exchange Binance hacked with over $5.6 billion stolen:
• $1.45 billion $BTC
• $875 million $ETH
• $234 million $TRX
• $50 million $XAUT
• $1.57 billion $USDT
• $823 million $USDC
• $75 million $WBETH
• $88.5 million $RLUSD
+ $447.5 million additional cryptocurrencies
This is the largest hack in crypto history.
Five Mistakes When Integrating AI into CRM and ERP 💩
The model may work perfectly in a test environment, but create SERIOUS problems once connected to a corporate system.
The reason is simple. In a production environment, AI interacts not only with data, but also with real business processes.
Mistake #1. Giving AI excessive access rights
If an agent only needs to read a client card to perform a task, it does not need permissions to modify the entire CRM database.
The principle of least privilege should also apply to AI.
Mistake #2. Connecting the LLM directly to critical operations
Creating payments, modifying contracts, deleting data, these are all actions that require strict validation and, in some cases, human confirmation.
Mistake #3. Ignoring the quality of source data
AI will not automatically fix contradictory reference data, duplicate clients, or outdated information.
In some scenarios, it will only make the consequences of these problems less noticeable.
Mistake #4. Not providing for action auditing
It is important to log which data the system used, which tools it called, and what changes it proposed or executed.
Mistake #5. Not designing for fault tolerance
What happens if the model is unavailable, returns an incorrect result, or the request limit is exceeded?
Mission-critical business processes must continue to operate.
Conclusion: A production AI solution is not an LLM with API access. It is a full-fledged information system with its own security architecture, monitoring, and operational rules.
Here is my question for CTOs...
Which of these risks do you consider the most underestimated? 🤔
#CRM #ERP #OODOO #SAP #AI #enterpriseai #ZOHO
🤖 Made with AI
1
200
Five Mistakes When Integrating AI into CRM and ERP 💩
The model may work perfectly in a test environment, but create SERIOUS problems once connected to a corporate system.
The reason is simple. In a production environment, AI interacts not only with data, but also with real business processes.
Mistake #1. Giving AI excessive access rights
If an agent only needs to read a client card to perform a task, it does not need permissions to modify the entire CRM database.
The principle of least privilege should also apply to AI.
Mistake #2. Connecting the LLM directly to critical operations
Creating payments, modifying contracts, deleting data, these are all actions that require strict validation and, in some cases, human confirmation.
Mistake #3. Ignoring the quality of source data
AI will not automatically fix contradictory reference data, duplicate clients, or outdated information.
In some scenarios, it will only make the consequences of these problems less noticeable.
Mistake #4. Not providing for action auditing
It is important to log which data the system used, which tools it called, and what changes it proposed or executed.
Mistake #5. Not designing for fault tolerance
What happens if the model is unavailable, returns an incorrect result, or the request limit is exceeded?
Mission-critical business processes must continue to operate.
Conclusion: A production AI solution is not an LLM with API access. It is a full-fledged information system with its own security architecture, monitoring, and operational rules.
Here is my question for CTOs...
Which of these risks do you consider the most underestimated? 🤔
#CRM #ERP #OODOO #SAP #AI #enterpriseai #ZOHO
🤖 Made with AI
45
Artem Techman retweeted
Unitree Introducing: H2-X Kung Fu Legend Guardian🤖
Powerful performance to serve and guard you, with a maximum endurance of approximately 8 hours.
Please use the robot safely and in a friendly manner.
Artem Techman retweeted
Debt of the biggest AI companies:
Amazon $AMZN: $132B
Oracle $ORCL: $125B
Alphabet $GOOGL: $123B
Meta $META: $84B
Broadcom $AVGO: $65B
Intel $INTC: $51B
Microsoft $MSFT: $40B
CoreWeave $CRWV: $36B
Dell $DELL: $35B
This is only bonds and loans.
Data center leases that have not started yet add $329B for Microsoft, $288B for Oracle and $279B for Meta.
Can the AI build keep running on debt?
By far the best ad I saw in the entire AI space...
Anthropic and Salesforce literally hired Matthew McConaughey to present their new MCP, which connects Salesforce and Claude, they called it Claudeforce.
The product in itself is pretty cool for mid-sized companies who work with Salesforce, but what I like is the simplicity in which they bring it to the audience.
Connecting your database and your tools to an AI is becoming a normal standard for every company now, but most people still don't fully get it.
Maybe ads like this will help people understand the productivity gains that hide behind connecting all your tools and having one central context brain for your business.
Anyways... another day, another new MCP and we just keep building.
For anyone who knows film please tell me how much the production of an ad like this costs?
1
104
OpenAI put more than 4 billion dollars behind a job title that Palantir invented in 2011. And it's by far the hottest tech job right now 🤯
Indeed (Job listing platform) listed 643 forward deployed engineer openings in April 2025. By April 2026 it listed 5,330, a 729% jump, per Business Insider's reporting on Indeed data.
An FDE is an engineer who works directly inside the customer's environment and carries AI integration projects A to Z. Same person runs discovery, data plumbing, permissions and production code. Palantir built the model for defence work, where software only counts once it runs inside someone else's perimeter.
On 11 May 2026 OpenAI launched the Deployment Company with 19 partners, and agreed to buy Tomoro, an Edinburgh firm whose 150 FDEs already shipped systems for Virgin Atlantic, Tesco and Supercell. Anthropic, Google and Salesforce hire the same profile. Salary trackers put median base near 190,000 dollars, with staff level packages above 630,000.
Why now: models stopped being the blockage. Integration, data quality and permissions is what really impacts the speed of deployment.
Have you hired your FDE yet?
Source: openai.com/index/openai-laun…
This robot learnt how to play baseball and threw a ball at 111km / hr 🤯 bruh...
Researchers at the Robotics and AI Institute have built AthenaZero, a robot that can throw, catch and hit balls at very high speed.
In one experiment it threw a tennis ball at 30.8 m/s, roughly 111 km/h. To make that possible, the robot was given unusually light and agile arms, while the heavy components were moved closer to the torso, which cut the effective mass of the limbs to about a tenth of that of a conventional robotic manipulator.
But the real story here is not baseball.
AthenaZero was built as a platform for fast physical actions, the kind where milliseconds genuinely matter. The robot can already throw, catch balls, hit them with a bat, and play catch with another robot and with a human. The researchers believe this approach could help create robots that interact with the world around them far faster and far more naturally.
It's only 2026 and robots already learnt how to play baseball, at this point I am really wondering what's next?
1
79
How AI vision models and AIR sort plastic in 2027. 🤯
At modern plants, bottles are not sorted by hand anymore, instead they use smart automation.
The process works like this:
1. Bottles travel along a conveyor, arranged in a single layer.
2. Optical sensors and infrared cameras identify the type of plastic (PET, HDPE) and its color in a fraction of a second.
3. At the end of the belt, a computer signals the air nozzles. A sharp blast of compressed air “shoots” the right bottle into its bin.
Do we really need humans for hard labour factory / plant jobs? Or should all of this be automated?
Follow @artem_techman to stay ahead in ai, robotics & tech.
61
An "ambulance" for humanoid robots has appeared in Japan 🤯
Japanese company GMO has launched a special vehicle that goes out to broken humanoid robots. Inside, there are an engineer, diagnostic equipment, tools, spare parts, and even a spare robot. If the fault can be fixed on the spot, the robot goes back to work. If not, the broken machine is taken in for servicing, and a working one is left in its place. The company calls this the first specialized transport for servicing humanoids in Japan.
For now, GMO has only one such vehicle in Tokyo, but the idea itself shows how a separate infrastructure is starting to form around robots, much like servicing and tow trucks emerged around cars. As humanoid robots become more widespread, the company plans to expand these services.
I think this is an absolutely genius and incredibly logical move by GMO Internet Group.
Follow Artem Gorelov to stay ahead in ai, robotics and tech.
Source: newatlas.com/ai-humanoids/ja…
66
Today we're launching skillsaggregator.com, a curated library of 450+ security checked skills for Claude. 🥳
New skills land on GitHub every week, and most people still spend hours digging through them and installing random ones just to test. That wastes time, and it opens the door to prompt injection and malicious code.
Today we fix that.
Skills Aggregator continuously scans GitHub for new skills, checks each one with NVIDIA's open source SkillSpector scanner, and lists the ones that pass. One trusted place for the best Claude skills.
Not sure where to start? Tell our Skills Aggregator AI what you're working on and it will recommend which skills to install first.
2 day free trial, then $99/year.
🎁 50% off for the first 100 users.
All Ryon.ai clients get free access for life.
👉 skillsaggregator.com
#claude #claudeskills #aiskills #mcp
AI turns an ordinary X-ray into 3D navigation for surgeons. Mindblowing...
MIT researchers have developed a system called xvr that matches a patient's X-ray images to a 3D model of that patient built beforehand from a CT or MRI scan.
The AI adapts to each individual patient in about five minutes, then performs the matching in a matter of seconds with sub-millimetre accuracy. The system can also generate around 1,000 synthetic X-ray images per second.
This could help surgeons see exactly where an instrument is inside the body during minimally invasive surgery. Today, a doctor has to work out the position of instruments relative to organs and bones in their head, using a flat X-ray. In tests on data from five hospitals, xvr proved significantly more accurate and robust than existing AI methods, and the researchers are already working on applying the technology to surgical robotics.
Source: news.mit.edu/2026/new-ai-tec…
82















