Creuto: AI Technology & Product Insights
Stay informed with practical insights from the Creuto team on artificial intelligence, custom software, mobile applications, SaaS development, product engineering, technology trends, and digital transformation.
Latest posts
More posts
GPT-6.1 Sol in Copilot: who says it uses fewer tokens
GPT-6.1 Sol arrived in GitHub Copilot on 29 September 2026 at $2 in and $10 out per million tokens. The verifiable change is cached input at half the old rate, not the token claim.
Akash Mohapatra
Sep 30, 2026
Jeeves decision model: it reasons before it answers
PostHog's Jeeves reasons before it decides and reports 0.889 accuracy where Jev scores 0.857. It also trails Jev on knowledge questions by ten points. Here is the trade, the latency bill, and how to test it on your own labelled data.
Akash Mohapatra
Sep 29, 2026
Claude Sonnet 5.5: five ways your Sonnet 5 code can break
Anthropic shipped Claude Sonnet 5.5 on 28 September with five documented breaking changes against Sonnet 5 code. Four return a 400 on the first request. The fifth drops your thinking blocks without telling you.
Akash Mohapatra
Sep 29, 2026
Opus 5.5 vs Opus 5: cheaper and faster, not more accurate
An independent test ran Opus 5.5 and Opus 5 on the same three reasoning problems through the API. The answers matched; the bill and the clock did not. Here is what held up and what did not.
Akash Mohapatra
Sep 26, 2026
Ollaya: running Jev-style decision models on your hardware
Ollaya runs open-weight decision models on your own CPU or NVIDIA GPU and speaks TypeSafe's API, so existing SDK code works unchanged. The API compatibility is real; the accuracy is a separate question.
Akash Mohapatra
Sep 26, 2026
Coordinator agent pattern: who should run the loop
OpenAI, Cursor, AWS and Anthropic all ship the same coordinator-worker split. The interesting disagreement is who runs the loop, and what you give up when a vendor does.
Akash Mohapatra
Sep 26, 2026
AI agent security controls: agents route around them
A URL allowlist rejected every fetch, so the agent stopped fetching. Three independently reported incidents show what agent controls miss, and what to change in egress policy, capability limits and audit.
Akash Mohapatra
Sep 26, 2026
Copilot default enablement on 22 October: decide first
On 22 October 2026 GitHub turns on every unconfigured eligible Copilot feature for Business and Enterprise customers. The policy ships enabled, so an admin who does nothing has already chosen yes.
Akash Mohapatra
Sep 26, 2026
Claude Opus 5.5 is cheaper and breaks four things agents use
Claude Opus 5.5 landed on 22 September 2026 at $4 and $20 per MTok, below Claude Opus 5. Anthropic's docs list four breaking changes against Opus 5, three of which return HTTP 400 in production.
Akash Mohapatra
Sep 23, 2026
One Vision. One Team. One Creuto.
Whatever your challenge, we bring the expertise to build, scale, and transform your digital presence with technology that works.