Ellie Daw (@ellieintech) header
Ellie Daw (@ellieintech) profile picture

Building https://nitter.cf/t.co/vsSQtBJXHU backed by Forum Ventures

San Francisco
Joined April 2017
Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)Photo by Ellie Daw (@ellieintech)
I had the opportunity to speak to the team at @kinandcarta about the challenges and opportunities of exposure notification tech during the pandemic. Their podcast captures some the important nuances and is definitely worth a listen! #privacy #covid19 #innovation
This quoted post is unavailable.
1
1
17
Starting is a company is just surviving long enough to turn learnings into value
9
241
Didn’t we used to say that premature optimization was the root of all evil in software? I feel like that’s where we are with integrating agents. How can we optimize something we don’t understand
1
4
151
More context is not always better for agents. They can anchor on weird facts and center their entire output on the wrong thing. The teams that feel good with their AI integrations are the ones focused on durable memory and context optimization
1
4
226
Everyone I talk to feels cognitive overload working with agents and I think it’s partly because the products were built to optimize the agent experience, not the human one
1
89
The stats in gjalla show codex usage starting to catch up to Claude Code
2
130
The coolest moments as a founder are findings the early adopters who agree with your view of the future and want to be a part of the journey to getting there
1
1
4
166
Agents don't really think about things the way we do... you may organize your docs & memory in a way that makes sense to you and your agents don't read it or retrieve it like you'd expect them to. People get frustrated but I'm on a mission to unfrustrate!
2
110
Ellie Daw retweeted
i just open sourced my autonomous job hunter project. the state of job hunting today is absolutely broken, so I made a tool which lets everyone automatically apply to jobs, hopefully pushing companies to adopt better processes by burning the system down github.com/sahiltalwar88/job…
1
1
2
92
good conferences are hard to find… @convex’s Abstract Conf feels like a whole swarm of folks with interesting thoughts and open minds
1
1
16
338
Don't get me wrong, I'm on the train where I think AI is extremely useful and has many interesting applications when applied the right way... but the first bullet point in the Fable announcement is so funny. Frontier intelligence that can finally... meet intern requirements
89
If you don't know about agent hooks, you need to. here's a quick rundown: A "hook" is kind of like a scheduled reminder for a coding agent. You can set up custom code or instructions to be run at a specific time in the agent's lifecycle. For example, maybe deleting a folder is something you always want an agent to ask you about. You could configure the "pre tool use" hook to check if the tool is a bash command and if it contains "rm -rf". If yes, escalate it to human review. This example shows how to check something before an agent uses a tool, but there are other times you can "hook" into the lifecycle... - at the beginning of a session - at the end of a session - at the beginning of a turn - at the end of a turn - when a subagent starts - before compaction - etc... Why am I telling you about this? A lot of teams building primarily with coding agents end up frustrated because their agents don't follow instructions, repeat the same mistakes over and over, and they're just playing whack-a-mole. This gets better the more deterministic you can make the workflow, and with coding agents that means using hooks. With it, you get token savings, more secure code, fewer frustrating incidents like "my agent deleted my entire home directory", the list goes on.
2
4
246
the conversation about shared memory seems to be picking up lately... part of why teams feel like their agentic coding workflows are like playing whack-a-mole is because they're using a bunch of tools that don't work together... your memory informing the hooks on your harness? :chefs_kiss:
1
4
155
things like this used to make me think "intern season" but now it's probably just "vibe code season"
95
one of the best things a customer has ever said to me is that their code progressively gets better and better the longer they use gjalla
4
150
over 14 billion tokens of sessions flowed through gjalla this week 👀 that’s the most yet
6
137
starting to see that memory pruning is one of the biggest levers to make your coding agents stop burning unnecessary tokens
74
the amount of token waste in any given agent session is insane...
1
2
136
no matter how many marketing emails I unsubscribe from, my inbound emails don't seem to calm down
7
135
And doing it alongside my TRIBE! Thanks for the shout out ✨🚀
Member spotlight: Ellie Daw. Ellie is a technical founder who has shipped more than 40 products from zero to one. She's building gjalla, making a company's technical infrastructure visible and taming tech debt, and runs Unstuck, her fractional technical-cofounder practice. This is who's in the room. What's the hardest zero-to-one problem you're working on right now?
10
1,185