@google is trying to sell its AI features via Google Workspace. Only one “recommended” plan, no option to change it in the admin console. I was always a loyal user but when I tried to set up a new account :(
@google you have been always an engineering company. This looks like a shitty SaaS business
TLDRL: UK freight transport companies continue to close across all sizes and all regions.
Simple statistics:
- 74% are small with < £100k in turnover
- 58% are one-man businesses
- 21% are loss-making
Large companies are not better off (20% of those with turnover >£10m are loss-making).
Bottom line: strong sales activity is critical for profitability at every size.
PS: Statistics were made from https://t.co/cFRTFM8FRm
UK–US trade is down, but some segments are still holding up
cars –£674m,
organic chemicals –£386m,
pharma –£309m,
scientific instruments –£60m,
aircraft +£67m,
power generators +£268m.
UK exports to the US fell by about £1.9bn this summer vs last year — the biggest drop so far. But it’s not across the board. A bunch of categories are still steady or even growing, which means logistics companies can shift focus instead of just absorbing the hit.
Sectors that are holding up: electrical goods, beverages and tobacco, office machinery, toiletries, textiles, furniture, fish and shellfish, food products, coffee/tea, cereals, paper and paperboard.
I use Cursor Auto mode, then Claude Code, then ChatGPT Codex — in that order.
Why? Mainly because of response speed, which gets worse in exactly that order.
I’m on the cheapest plan for all of them, and I use them inside Cursor.
Another big thing: Cursor indexes the whole codebase, while Code and Codex just do search.
Also, all these companies roll out model updates or switch to quantized versions (worse quality) whenever there’s heavy load.
That’s why I don’t buy into the “Claude is better than Cursor” talk. Today maybe, tomorrow maybe not.
My algorithm is simple: I go down the list of models, and if after three tries the current one hasn’t given me anything useful, I switch to the next.
I’ve been writing 100% of my code with LLM for the past two weeks.
Wrote and refactored thousands of lines in Cursor using $20 auto models + CLINE with Google CLI.
I think I won’t write code by myself anymore, no point in it.
I get very tired from sessions with Cursor, but I enjoy more writing texts with development rules for it.
Later I will write recipes.
For now I’m thinking to buy premium Cursor or Claude Code for $200.
Even current setup is fully enough, but I want faster!
Our custom AI deployed! A new summary for a small subset of companies.
See an example here:
https://t.co/LlwgFjzWQm
It was cooked on our local AI machine with open-sourced LLMs.
Right now, we are deploying a new search algorithm!
Tweaking some internals significantly improves search quality.
Before:
The old search returned relevant results for "Francesco Blasi" but not for "Blasi Francesco." Now, it works perfectly for both variants.
We rolled out exporter and importer profile patterns for UK exporters and importers.
Just one glance is enough to determine what type of goods they are dealing with!
Here is the example of @UNIQLO_UK
https://t.co/3VCW63LkZn
We are rolling out a new design for company pages!
Badges for a quick assessment of any UK company:
- Company size (Large, ...)
- Company status
- Years incorporated
- Last reported revenue, employees, and assets if reported
Green colors for good indicators, red for the bad ones