AI in Software Engineering

Keeping an enterprise AI stack current — and why teams fall behind

Upgrade debt accumulates quietly until a CVE makes it urgent. A candid look at why it happens, using our own estate as the example.

AdminFounder & Engineering Lead · August 31, 2026 · 8 min read

Keeping an enterprise AI stack current — and why teams fall behind

In the week of 25 July 2026, Next.js shipped 16.2.12, Payload CMS shipped 3.86.0, LangChain core reached 1.5.2 and pgvector 0.8.5. Keeping pace with that cadence is not a matter of discipline alone — it is a matter of having a mechanism.

How upgrade debt accumulates

Coupled dependencies that cannot be upgraded independently
No routine signal, so drift is only noticed during an incident
Fear of breaking a working system, which grows with the size of the gap
Upgrades competing with feature work for the same capacity

A candid example

This site ran an unpatched Next.js version through the React Server Components deserialisation RCE in 2026 — an unauthenticated, single-request path to arbitrary command execution.

The mechanism that actually fixes it

Sources

Next.js releases — https://github.com/vercel/next.js/releases/tag/v16.2.12
Payload CMS releases — https://github.com/payloadcms/payload/releases/tag/v3.86.0

Keep reading

All articles

AI Strategic Insights

Getting cited by AI assistants: answer engine optimisation for B2B technology buyers

73% of B2B buyers used an AI answer engine during vendor evaluation in 2026, and brands cited in the top three positions saw materially more qualified demand. AEO is now part of how buyers find you.

August 29, 2026· 7 min read

AI Strategy

Time to value: when enterprise AI actually pays back

Median time-to-value on agent deployments in 2026 is around 5.1 months, but the spread by function is wide — from 3.4 months for sales development to 8.9 for finance and operations.

August 26, 2026· 6 min read

Retrieval & RAGKnowledge AI

What enterprise RAG actually has to get right

Retrieval is now described as the foundation tier that grounds agents in real enterprise data. Getting it right is mostly about retrieval quality, permission-aware filtering, and knowing when to refuse.

August 22, 2026· 8 min read