Remember all the headlines about Claude replacing COBOL developers and IBM's stock plunging after Anthropic's announcement?
With all the hype surrounding AI agents replacing COBOL modernization, and Anthropic recently hosting a webinar demonstrating its capabilities, we felt it was important to go beyond the headlines and put these claims through a rigorous technical evaluation.
What we discovered told a very different story.
After deploying 222 AI agents, the plugin failed to convert the COBOL program we gave it. It failed 61 of 62 modernization requirements, silently altered business logic, and reported success when critical functionality was broken.
By comparison, Codegiant successfully met all 62 requirements using our purpose-built Kronos LLMs and deterministic modernization system.
The bottom line? In our testing, Anthropic's plugin simply didn't work as a viable mainframe modernization solution.
We published our detailed findings here:
buff.ly/rPtNopH
The AI model layer is turning into a compression chain.
Massive raw data → US frontier models.
US frontier models → Chinese frontier models.
Now the rest of the world will use cheaper Chinese frontier models as the base layer for fine-tunes, SLMs, and specialized private models.
Solutions Engineers are being rebranded as Forward Deployed Engineers.
✨ Shiny new title. Same enterprise reality.
Frontier models do not magically deploy into the enterprise.
You still need layers of software, integrations, workflows, permissions, observability, security, governance, and consulting just to make them useful in production.
That is the cost of using a generalist model for everything.
🧰 Jack of all trades. Master of none.
My feed is filled with Claude Code, Codex, and Cursor posts.
But generating code was never the hard part.
Deploying to the enterprise is.
We're incredibly excited to be working with our partners in Japan on mission-critical systems that have a meaningful impact at scale. 🇯🇵
If you want a cost-effective, game-changing way to build software, watch this: youtube.com/watch?v=mayKAsAc…
Small language models (SLMs) with the right abstractions will drop token costs fast.
Frontier AI companies know this.
That’s why they are racing to own the tooling layer around the model:
Infrastructure. Developer tools. Deployment. Orchestration.
If models get smaller and cheaper, the money moves to the system around them.
OpenAI, Anthropic, and the other frontier companies will become SaaS 3.0.
At Codegiant, we spent the time to build the systems frontier companies are racing to acquire.
I'm waiting for my feed to fill up with Claude Fable 5 posts about how it revolutionized opening Gmail and handling deterministic actions that could have been solved with basic automations.
I'm off to Japan to work with a customer, and I was going to post something inspirational about running an LLM on my laptop during the flight.
Then I thought to myself: thank God there are no AI agents, Claude Code projects, or YC-funded software powering any aspect of this plane.
Rishi Mathur retweeted
🚀 From idea to mvp in 6 weeks! 🚀 team Yareta built a scalable mvp and secured $100k in contracts with codegiant’s one-click solution and @Cloudflare highly secured and scalable Infrastructure. See how they saved time and costs! Read more: buff.ly/3AEqhPH
Rishi Mathur retweeted
Statistics for Codegiant 3.0, look how your product has grown on @ProductHunt.
Good job @rishikmathur
You can see more here - upvote-bell.com/projects
#foodpanda text message verification and customer support doesn’t seem to work? Technical difficulties?