Christian Gambardella (@heygambo) profile picture

Software engineer, AI agents in my daily work.

Berlin, Germany
Joined January 2009
Photo by Christian Gambardella (@heygambo)Photo by Christian Gambardella (@heygambo)Photo by Christian Gambardella (@heygambo)Photo by Christian Gambardella (@heygambo)Photo by Christian Gambardella (@heygambo)Photo by Christian Gambardella (@heygambo)Photo by Christian Gambardella (@heygambo)Photo by Christian Gambardella (@heygambo)Photo by Christian Gambardella (@heygambo)
Who is still using Algolia? Curious why
16
already wishing a happy christmas to the cashiers 🌲
19
Christian Gambardella retweeted
Protesting an EU-INC company law that doesn’t even exist yet because it might have downsides is peak European self-sabotage. You want higher wages? Make Europe the best place in the world to start and grow a business. European founders still face 27 systems, endless friction and brutal regulations – and too few reasons to build here instead of in the US. These people spend their working hours and money fighting a potential reform. Try building a business here and see how hard it is. We need an economy where successful businesses can grow – and workers can participate in that success (via higher wages AND stock options). #EUInc
30
21
4
299
15,394
Christian Gambardella retweeted
There are only five levels of working with agents: Level 1 - you paste the context Level 2 - you save your prompts Level 3 - the agent pulls context itself Level 4 - your team runs the same setup Level 5 - the setup catches what you forget Most people stop at 2.
1
1
23
There are only five levels of working with agents: Level 1 - you paste the context Level 2 - you save your prompts Level 3 - the agent pulls context itself Level 4 - your team runs the same setup Level 5 - the setup catches what you forget Most people stop at 2.
1
1
23
Christian Gambardella retweeted
Mistakes I made putting agents into real work: - Assumed a skill deploys what it builds - Expected the skill to tell me where it stops - Mixed three topics into one chat The first one cost me a deadline.
1
1
54
Mistakes I made putting agents into real work: - Assumed a skill deploys what it builds - Expected the skill to tell me where it stops - Mixed three topics into one chat The first one cost me a deadline.
1
1
54
Christian Gambardella retweeted
2024: Don't let an agent near production without reading every line it wrote. 2026: Don't go near production without asking an agent what else runs at that hour.
1
15
2024: Don't let an agent near production without reading every line it wrote. 2026: Don't go near production without asking an agent what else runs at that hour.
1
15
Christian Gambardella retweeted
Some mistakes you can't work your way out of. If the thing takes longer than the time you have left, effort stops mattering. No amount of panic shortens it. Worth knowing which of your steps are like that, before an agent takes the rest off your hands.
1
11
Some mistakes you can't work your way out of. If the thing takes longer than the time you have left, effort stops mattering. No amount of panic shortens it. Worth knowing which of your steps are like that, before an agent takes the rest off your hands.
1
11
Christian Gambardella retweeted
What people use agents for: - Writing code - Fixing tests - Refactoring What pays off more in an old system: - Mapping what runs when - Finding a safe window to deploy - Explaining why something is slow Reading beats writing here.
1
11
What people use agents for: - Writing code - Fixing tests - Refactoring What pays off more in an old system: - Mapping what runs when - Finding a safe window to deploy - Explaining why something is slow Reading beats writing here.
1
11
Putting agents into a team is an organizational process disguised as a technical one.
8
Christian Gambardella retweeted
What an agent can't find in an old codebase: - Why the workaround is still there - Which service nobody touches on Fridays - The naming rule four people agree on - What broke last time someone tried this None of it is written down. That's the work.
1
1
13
What an agent can't find in an old codebase: - Why the workaround is still there - Which service nobody touches on Fridays - The naming rule four people agree on - What broke last time someone tried this None of it is written down. That's the work.
1
1
13
Christian Gambardella retweeted
The fastest way to make an agent useful in a large codebase is to write down the rules nobody ever wrote down.
1
11