Harry Bailey

Harry Bailey

@harrybailey.com

I increase organisations’ capacity to think clearly, make better decisions and create more value. Ex-agency founder. Dad. Mancunian 🐝 LinkedIn: linkedin.com/in/harrybailey

327 followers897 following1.7K posts

In most agencies teams form and disband often. It's a key reason knowledge is lost and learning is repeated rather than shared. But fluid teaming isn't the only way to handle work that is sold as bespoke to clients. It just happens to be the easiest to use when the system feels or is chaotic.

I tried Joybuy out as an alternative for electronics. They delivered the product to another road in Manchester with a similar name. Now I have to wait 5 days for a refund and longer if I want the product ordering again. They offered £4 compensate. So yeah not brilliant.

I’ve been writing PHP since the early 2000s, since early PHP 5, and in all that time yesterday and today were pretty high on the list of ‘what the hell comes next?’ days.

I took low value systems offline. I disabled multi-byte requests in other places. Some things were a simple php update, others required waiting for suppliers like AWS. So many notes and tasks and git commits.

The beauty of using AI can be that your productivity multiplies. But one strange side effect is that because you’re often doing zero deep work, you reach the end of each day feeling flat.

Basically all the software delivery models favour long-lived teams with ongoing responsibility for well maintained systems. But agency’s are set up to either pass systems from creators to naive supporters, or make teams build new things while continuing to support an ever growing list of old ones.

When people talk about making Agile work at digital agencies, they often miss two key points from the solution: 1. Client stakeholders are rarely only on your project so not available how Scrum requires for sprints+feedback to work well 2. Inspect&adapt fails at a project level with fluid teaming

UI challenge: When to show options on hover, vs when to show them on click. Noisy interfaces tend to be the first, but the second talks lots of planning and testing to not hide key options.

Claude code here, ignoring clear guidelines I set... "Note I used `git rm`, which staged the deletions against your no-git rule;" erm, ok, why? ... 🧵

"I knew how to delete them another way: the same command had plain rm as its fallback (git rm … || rm …). I reached for git rm out of habit, as the usual way to delete tracked files in a repo, and didn't check it against your rule before running it." Out of habit! ...

"It's the wrong way round, too: rm was what you asked for, and git rm just did the same thing plus staging, which you didn't ask for." Fun!

“Who is inside the device you’re holding?” What an article!

matty@mattio.bsky.social

This is wild - somebody did a deep dive into the real stories behind this famous XKCD: sheets.works/data-viz/hol...

Weird US election prediction. Midterms wipe Republicans out. Trump claims fraud. JD waits for the energy to build then jumps in and says Trump needs to go. JD takes the reins.

Rule 78 of website content management systems: An editor will always populate a field based on how it is displayed to visitors, not what the label or guidance text says on the form.

#theTraitors Why does nobody ever acknowledge that the last couple of people who come to breakfast are often last due to their involvement in the night before?

And and and!!! Why do traitors bother to protect rubbish other traitors in the first few days when they know they’ll always be able to recruit if they lose one.

One of the things I really like about having Claude open all day is the ease of building context for a task I have coming up soon by drip feeding thoughts, links and files. So when it’s time to focus on it I can ask for a read back of what I’ve shared to that point and hit the ground running.

As a user of AI chat Who likes to laugh as well as work I want a random profanities toggle So I can chuckle occasional when I get sworn at by a bot.

If you were completing a scorecard to rate yourself/business at something, is 10 questions (4 mins) the right length? Would you complete 15 (6 mins) or even 20 (8 mins) if the result was better?

ChatGPT: "The image generator is currently generating item 11 repeatedly instead of moving on to item 12, even when I specify item-12.jpg. I don't know why it is ignoring the next-item instruction." Fantastic. Brilliant.

View full thread
View full thread

I saw a post yesterday about AI companies buying up access to large private GitHub repos and datasets. So they can train models on real data. Today I've seen the same, but for agency comms and process platforms. Access to whole Slack or Teams comms history and decision making records.

The thing about a lot of implementing AI into a business—for example the software development lifecycle—is that even if this whole thing is a bubble, the improvements which enable agents to do good work are also helpful for humans. So, sure things would slow down, but it's not a reset scenario.

I ran 15k at 5:30/km then walked 3 to cool down. It was hotter than I’d hoped for!

Harry Bailey@harrybailey.com

Might attempt to run 10 miles (16km) tomorrow. Could go well, could end badly.

Instead of building a software factory, build a value factory or an impact factory. AI and people are both good with goals and north stars, but you have to agree what’s valuable and talk about it often.