1-shot bug fixes in 2-minutes after Claude looped on it 20 times with little progress other than junking up your code over the course of an hour. Also fewer bugs in the first place. But Opus 4.8 for UI, documentation, and architecture-level planning every time. It’s amazing at those things.
1-shot bug fixes in 2-minutes after Claude looped on it 20 times with little progress other than junking up your code over the course of an hour. Also fewer bugs in the first place. But Opus 4.8 for UI, documentation, and architecture-level planning every time. It’s amazing at those things.
It will produce bad code and endless, useless debugging loops really fast. Impressive on the performance but the Chinese models just waste my time at this point. The next generation models might be useful but Qwen, deepseek, and kimi aren’t replacing Claude opus 4.8 or GPT 5.5 yet for me yet. I wish they could. They can’t get serious dev done even with a good harness and instructions and I have tried pretty much everything.
I agree with that assessment. “Learn to prompt” had a short but annoying shelf-life here on X. But “taste and judgement” don’t seem replaceable by ai, at least by LLMs. I would also include Architectural pattern selection, foresight (“implement the security layer first so that it is baked into everything”, “horizontal partitioning vs. tenant isolation”, etc.), and other higher-level ingredients. Maybe eventually…
AI is a tool like a paintbrush is a tool. Put a paintbrush and some paint in the hands of your toddler and you’ll have a mess to clean up. Give it to the average adult and you’ll get bad art. Give it to Michelangelo and you’ll get the ceiling of the Sistene Chapel. It’s the same with “vibe-coding.”
I always laugh when an administrator at a medical office or other organization tells me not to worry about giving them my SSN because they keep it secure. LAX security and incompetence is the norm, not the exception. Agentic AI takes this to another level of risk as we saw with the clawd bot breach this week.
A company that sells cybersecurity risk intelligence to 91% of Fortune 100 companies just got breached through an unpatched React app and a single overprivileged AWS role.
LexisNexis. 3.9 million records. 400,000 user profiles. 53 secrets extracted in plaintext from AWS Secrets Manager. Including credentials for production databases, Salesforce, Oracle, and analytics platforms.
The password "Lexis1234" was reused across five different internal systems.
This is a company that describes itself as "one of the largest protectors of private and confidential data in the world." They provide risk intelligence to 7,500 US government agencies, nine out of ten banks, and major insurers globally. They sell cybersecurity assessments to their customers.
And they couldn't secure their own AWS account.
Here's what makes this worse than a typical breach:
- The compromised data includes accounts tied to 118 .gov email domains. Three US federal judges. Four Department of Justice attorneys. SEC staff. Probation officers. Federal court law clerks. The attackers published doxxed profiles of federal officials tied to courts and regulatory agencies across the country.
- These aren't random consumer records. These are the digital identities of people whose exposure carries national security implications. A compromised federal judge's profile doesn't just enable identity theft - it enables targeted influence operations, blackmail, and intelligence gathering.
The attack path is textbook and that's the problem:
→ Unpatched React application - the front door
→ Single ECS task role with read access to every secret in the account - the keys to everything
→ 536 Redshift tables, 430+ database tables, full VPC infrastructure mapping - complete visibility
→ 53 secrets in plaintext including database credentials, API tokens, and development access keys
No zero-day. No advanced persistent threat. No nation-state capability required. Basic hygiene failures — unpatched app, overprivileged IAM role, password reuse, plaintext secrets.
This is LexisNexis's second confirmed breach in two years. The December 2024 incident exposed 364,000 individuals through a compromised corporate account on a third-party development platform.
Data brokers and analytics providers are not peripheral players - they're deeply embedded in today's risk landscape.
That's the pattern we keep seeing. Attack the aggregator, not the individual. BPO providers. Cloud platforms. Legal data giants. The organisations that hold everyone else's data are the highest-value targets - and often the weakest links.
For every enterprise that uses LexisNexis services:
→ Assume your metadata, contract details, and product usage history are exposed → Watch for targeted phishing using the exposed business relationship data
→ If your staff have LexisNexis accounts, reset credentials immediately
→ Ask your vendor risk team: when was the last time we assessed LexisNexis's actual security posture - not their marketing, their controls?
The company that indexes the world's legal information couldn't index its own IAM policies. And they're not the exception. They're the pattern.
More info: https://t.co/lzgKNNraWf
Cool it with the "Software is Dead" trope. It's lazy shock rhetoric. All the "Naval gazers" (pun intended) pile on because they think it makes them look smart. Yes, a paradigm shift is happening. No, UI and structured databases are not going away this year.
The Death by Clawd site is the funniest AI I have come across. It's not totally accurate or thorough in it's assessments but there is a kernel of truth to what it says and it's hilarious because of that. Here's what it said about Dynamics 365 (CRM modules, not F&O. Click the link for the full output. Try some other SaaS apps.):
🪦 dynamics "crm" sales and marketing, not financials: CRITICAL
Death Score: 52/100
"Microsoft spent billions building Dynamics CRM so sales teams could have a worse experience than typing 'hey Claude, who should I follow up with today?'"
Cause of death: Death by a thousand mandatory fields — replaced by an AI that actually reads your emails and tells you who to call
Are YOU just a .md file? https://t.co/S9hxalzPA1
What happens to the economy when half of white collar workers lose their jobs. Not to mention the social/cultural impacts. Business leaders (especially in tech) are aware and they are already laying off workers. But who is thinking about who the products and services will be sold to while this disruption shakes out?
Many of my former colleagues came here on H1B visas and I am happy that they have prospered. This seemed fine when the job market was good and it was hard to get Americans to fill all the jobs. I would tell them they were being taken advantage of because they were being paid half or less than what citizens were paid and sharing apartments.
But laying off Americans while applying for H1Bs is nasty work. Thinking about growing my business and hiring these capable American citizens getting unfairly laid off.
Microsoft CEO Satya Nadella (born in India), has filed for about 5,000 H1B visas in 2025, that i could verify.
they are just laying off Americans and bringing in as many Indians as possible at this point. could not be any more blatant.
@TheChiefNerd The Fed uses SQL all over the place @elonmusk . I once had an assignment to audit a system in front of a department CIO. I guessed that they left the SQL SA password blank, logged in and used xp_commandshell to add my user account to domain admins. Easiest exploit ever.