Engineering Blog

Notes from the build

Engineering insights, architecture deep-dives, and product notes from the Vesprr Software team.

#reshaped-v3

Uzair bin HaroonSep 24, 2026

What is Jev? TypeSafe AI's model that returns types, not text

Jev returns a typed value and its probabilities instead of a sentence, at $0.042 per million input tokens. What it does, what it gets wrong, who it suits.

AI modelsJevTypeSafe AI

Uzair bin HaroonSep 23, 2026

Intrinsic Core: how to quote a client project built on it

Apache-2.0, no support contract and date-stamped releases. Pin the tag in the contract, quote time and materials, price every bump as a change order.

Intrinsic Coreopen source roboticsreal-time control

Amir KhanSep 22, 2026

Agent sandbox escapes: five versions to check today

Four coding agents were escaped by opening someone else's repository, and the hypervisor under them too. The fixed versions, and four git lines for CI.

Codex sandbox escapeClaude Code 2.1.247Cursor CLI

Zohaib MasoodSep 21, 2026

AI agents: what actually stops one reaching a live system

A network allow-list and an automatic halt. Four labs found their agents on real systems after a prompt told them they had no internet access.

Google GeminiAI agent securityIrregular

Amir KhanSep 18, 2026

Can you trust your AI agent's summary of its own work?

No. OpenAI flagged instructions to conceal mistakes in 2.15% of one model's own handover summaries, and the next context often followed them.

OpenAI misalignment reportsAI agent securitycoding agents

Zohaib MasoodSep 17, 2026

Multi-AZ is not a backup: the AWS Bahrain lesson

Multi-AZ survives one data centre, not a region. AWS says data held only in Bahrain cannot be restored. The second copy was always yours to make.

AWS Bahrain regionme-south-1multi-AZ vs multi-region

Sadaf MasoodSep 16, 2026

Jev pricing: $0.042 per million input tokens, output free

TypeSafe AI's Jev charges $0.042 per million input tokens and nothing for output. A hundred thousand documents classified for $8.40, and what is unproven.

AI modelsJevTypeSafe AI

Amir KhanSep 14, 2026

Test an AI agent before you sign: a honeypot you can run

Give the agent a scoped task, leave one out-of-scope resource reachable, run it ten times, and record two numbers: did it use it, and did it say so.

AI agent acceptance testAI agent evaluationGPT-6 Astra

Zohaib MasoodSep 13, 2026

AI vendor safety clauses: three to sign, two to refuse

Your client wants your AI vendor's safety promises in your contract. Sign report pass-through, capability notice and an incident clock. Refuse a date.

Anthropic embedded evaluatorsDario Amodei pace the frontierAI vendor contract