The thing you built on has an end date
A widely self-hosted automation tool announced it is archiving itself this month, and a model a lot of automations run on gets 50% dearer on 1 September. Both are the same question: what happens to your process when the layer underneath it moves?
- 01
Flowise is shutting down and archiving the repo on 10 August
The team behind Flowise, the drag-and-drop tool a lot of small firms use to wire up AI workflows, has called time on it: code freeze was 29 July, the repository gets archived on 10 August with packages marked deprecated, and they step back from community support channels on 31 August. The reasoning they give is that people increasingly hand this kind of work to coding agents instead. The Apache-2.0 source stays on GitHub and they invite forks, so nothing stops working overnight — but nobody is shipping fixes to it after that date.
FlowiseAI - 02
Claude Sonnet 5 introductory pricing ends 31 August
Anthropic's pricing page states the introductory rate of $2 per million input tokens and $10 output runs through 31 August, with standard pricing of $3 and $15 taking effect on 1 September. The same page notes that newer models use a tokenizer producing roughly 30% more tokens for the same text than the previous one. If a developer priced your automation against the current rate, ask them to re-run the sums before September rather than after the invoice.
Claude Platform docs - 03
The Rust project writes down when contributors may use an LLM
One of the largest open-source projects in the world published a policy on AI use, and the line it draws is fine to use a model to analyse, check, review or refine, not to create. Contributors must disclose LLM involvement, generated code is held to a higher standard than hand-written, and pasting model output at a reviewer without understanding it is grounds for closing the pull request. Worth reading as a template: it is a page long, it is specific, and most firms letting staff use AI still have nothing written down at all.
Inside Rust Blog - 04
INTERPOL links AI to 55% of reported cybercrime in Africa
The African Cyberthreat Assessment Report 2026, published on 3 August and drawn from 36 member countries, puts AI behind 55% of reported cybercrime in the region, with losses more than doubling since 2024 from USD 192 million to USD 484 million. The pattern it describes is automation across the whole chain: reconnaissance, phishing, social engineering and extortion. The geography is not ours, but the tooling is the same tooling, and business email compromise is named as one of the prolific categories.
INTERPOL - 05
Cloudflare opens wallets that AI agents can spend from
Announced on 4 August: programmable wallets giving an agent its own identity and the ability to pay for APIs, tools and content over the x402 protocol, with spending caps, allowlists and maximum transaction sizes set by the human who owns the account. You can claim a handle now; funding and actual payments are described as coming soon. File under things to watch rather than things to do — but the direction is software that buys things on your behalf, and the control you get is the one you configure up front.
Cloudflare - 06
Comments now pin to a moment in a Google Drive video
Comments on video files in Drive can be anchored to a timestamp and appear as markers on the timeline, so a reviewer can jump straight to the moment being discussed. It is available to all Workspace customers, Workspace Individual subscribers and personal Google accounts, it is on by default with no admin control, and comments can only be created on the web. Small, free, and immediately useful if anyone in your firm records a walkthrough on their phone and sends it for comment.
Google Workspace Updates - 07
Simon Willison's LLM tool grows reasoning traces and server-side tools
LLM 0.32 landed on 4 August, which Willison calls the most significant release since the project started: visible reasoning traces sent to standard error, support for providers' own server-side web search and code execution, and a rebuilt content-addressable log store. He notes it is turning into an agent framework, including tool chains that pause for a human to approve a step before continuing. Relevant if your developer builds you anything scripted — this is the free command-line end of the market catching up with the paid platforms.
Simon Willison
The Flowise post is the one I'd read twice, and not because Flowise matters.
If you run a small firm and someone set you up with AI automation in the last two years, there is a decent chance this is the kind of thing it runs on — a box with boxes and arrows on it, wired up in an afternoon, quietly moving your enquiries around ever since. On 10 August the repository gets archived. Nothing breaks that day. The code is Apache-2.0 and stays on GitHub, forks are welcome. But the queue of people fixing it when a provider changes an API drops to whoever volunteers, and that is a different arrangement from the one you thought you had.
Put it next to the Sonnet price change and it's the same story told twice in a week. The tool gets archived. The rate goes up 50% on a date already printed on the pricing page. Neither is anyone behaving badly, and neither is your fault. It's just that the layer under your process moves, and it doesn't ring to tell you.
I got used to this on site with a different vocabulary. You don't specify a fan because a rep is friendly; you specify it because you can still get parts for it in five years. The install is the cheap bit. The decade after it is where the money goes.
So the question I'd put to whoever built your automation isn't "is it working". It's what happens to it if the thing it sits on stops being maintained — and how many hours that answer costs. If nobody can tell you, you haven't bought a system. You've borrowed one.
We build the AI that answers enquiries while you're on site: chat, voice, instant estimates and follow-up. See how it works or price it in two minutes.