Ajaya

1.4K posts

Ajaya

Ajaya

@aj_agr

Engineer - building https://t.co/D4he05PfYi. Staff@anti-complexity lobby.

Bengaluru, India Katılım Nisan 2012
1.8K Takip Edilen213 Takipçiler
Ajaya
Ajaya@aj_agr·
"Possible cybersecurity risk." On my own system. You don't have the context to decide that — this is just unwarranted policing dressed up as safety. @OpenAI @thsottiaux
English
0
0
0
12
Ajaya
Ajaya@aj_agr·
Loving gpt-5.5 - a great model @sama - not freaking chatty like 5.4 - feels super smart as well - just gets it. Caught a bunch of errors made by opus-4.7 - fixed those - thought through edge cases before I did. :)
English
0
0
0
39
Ajaya
Ajaya@aj_agr·
Eventually this is everyone in the company.
Aaron Levie@levie

The more enterprises I talk to about AI agent transformation, the more it’s clear that there is going to be a new type of role in most enterprises going forward. The job is to be the agent deployer and manager in teams. Here’s the rough JD: This person will need to figure out what are the highest leverage set of workflows on a team are (either existing or new ones) where agents can actually drive significantly more value for the team and company. In general, it’s going to be in areas where if you threw compute (in the form of agents) at a task you could either execute it 100X faster or do it 100X more times than before. Examples would be processing orders of magnitude more leads to hand them off to reps with extra customer signal, automating a contracting review and intake process, streamlining a client onboarding process to reduce as many straps as possible, setting up knowledge bases than the whole company taps into, and so on. This person’s job is to figure out what the future state workflow needs to look like to drive this new form of automation, and how to connect up the various existing or new systems in such a way that this can be fulfilled. The gnarly part of the work is mapping structured and unstructured data flows, figuring out the ideal workflow, getting the agent the context it needs to do the work properly, figuring out where the human interfaces with the agent and at what steps, manages evals and reviews after any major model or data change, and runs and manages the agents on an ongoing basis tracking KPIs, and so on. The person must be good at mapping the process and understanding where the value could be unlocked and be relatively technical, and has full autonomy to connect up business systems and drive automation. This means they’re comfortable with skills, MCP, CLIs, and so on, and the company believes it’s safe for them to do so. But also great operationally and at business. It may be an existing person repositioned, or a totally net new person in the company. There will likely need to be one or more of these people on every team, so it’s not a centralized role per se. It may rile up into IT or an AI team, or live in the function and just have checkpoints with a central function. This would also be a fantastic job for next gen hires who are leaning into AI, and are technical, to be able to go into. And for anyone concerned about engineers in the future, this will be an obvious area for these skills as well.

English
0
0
0
36
Ajaya
Ajaya@aj_agr·
I just don't get the hype with Mythos - it's the same old story of hyping up a model before release. - Remember how agi was already achieved last year itself. - The marketing with software security looks sketchy to me tbh.
English
0
0
0
20
Ajaya
Ajaya@aj_agr·
I have switched back to 5.3-codex as my daily driver - 5.4 is chatty af - dumps a bunch of information where it's hard to figure out signal vs noise. Opus 4.6 for product/UX/UI work.
English
0
0
0
48
Ajaya
Ajaya@aj_agr·
If you've never worked with .jsonl or ndjson files, sorry but you're not an ai native engineer yet. 🤓🤓🤓
English
0
0
1
27
Ajaya
Ajaya@aj_agr·
Who is building stackoverflow/discord for coding agents to hang out on? It could be a very interesting market place for smaller/cheaper agents to just pay some money and get answers from expert/bigger models.
English
0
0
0
50
Ajaya
Ajaya@aj_agr·
@badlogicgames @baanditeagle I would love to help in development or early testing/feedback - please let me know. I am open to sponsorship as well if it helps. Thank you for Pi. :)
English
1
0
1
33
Mario Zechner
Mario Zechner@badlogicgames·
People of pi. I'm going to break the extension API hard. Specifically, business logic (event handlers, custom tools/compaction/etc.) needs to be split off from the ui layer. it will likely not be a massive amount of work to migrate an existing extension, but it will hurt a little.
English
44
9
551
66.4K
Ajaya
Ajaya@aj_agr·
@bleuonbase @badlogicgames @baanditeagle @southpolesteve @bleuonbase - that's awesome. I haven't tried out opencode yet - we are all-in on Pi - heck it's even replaced cc/codex in my local. If you have the code available, would love to take a look - I am sure we can exchange/share ideas - the fundamental problem is the same.
English
0
0
2
48
agusti
agusti@bleuonbase·
@aj_agr @badlogicgames @baanditeagle I forked @southpolesteve opencode-do example and added sse streaming and event-ids so client/server can reconnect and pick up where they left off (cf workers drives the servers session and it sync when u attach opencode tui
English
1
0
3
69
Ajaya
Ajaya@aj_agr·
@badlogicgames @baanditeagle This is exactly what I built custom on top of pi using an extension. Not super ergonomic of course - server mode may help get rid of this custom code. If I may request, letting clients send a idempotency key along with a user message would be awesome.
English
1
0
2
90
Mario Zechner
Mario Zechner@badlogicgames·
@aj_agr @baanditeagle servar handles sessions, clients gets latest session state on resume, then all subsequent updates.
English
2
0
5
568
Ajaya
Ajaya@aj_agr·
@badlogicgames @baanditeagle How does the client reattach to the old request and start streaming from where it left off? Is the sessionl file enough to make it happen?
English
1
0
0
532
Mario Zechner
Mario Zechner@badlogicgames·
@aj_agr @baanditeagle server handles agentic loop. clients just observe it. of a client disconnects, server keeps the agentic loop alive.
English
1
0
5
709
Ajaya
Ajaya@aj_agr·
@badlogicgames @baanditeagle Imagine you have a client which can disconnect mid stream - a browser for example. How do you reliably keep processing a request while the client can reconnect again? Maybe I have over engineered my solution - but this is a classic coding agent in a sandbox kinda thing.
English
1
0
0
734
Mario Zechner
Mario Zechner@badlogicgames·
@aj_agr @baanditeagle i don't know what you mean by "store and stream love events". server mode will be codex app server on steroids. local server managing multiple sessions clients (tui or web,) can connect to.
English
3
1
24
4.1K
Ajaya
Ajaya@aj_agr·
@badlogicgames @baanditeagle How is the server mode different from the current rpc mode? We have had to write an extension to store and stream live events from the pi rpc mode. Does the new server mode solve any of that?
English
1
0
1
1K
Mario Zechner
Mario Zechner@badlogicgames·
@baanditeagle more abilities. legacy extensions will continue to work normally. they will only need migration if you want them to work with the new server mode i'm cooking.
English
3
0
32
3.4K
Ajaya
Ajaya@aj_agr·
If Delve could hand out SOC 2 reports like this, then who allowed Delve to hand out SOC 2 like this? What exactly does a SOC 2 report guarantee anymore? And why assume others are better?
English
0
0
0
72
Ajaya
Ajaya@aj_agr·
openai cooked with gpt-5.4 - hard BE problems go to gpt 5.4 - FE gets built by Opus 4.6 now.
English
0
0
2
68
Ajaya retweetledi
Sowmya Ranganathan
Sowmya Ranganathan@heysowmya·
We've been shipping enterprise AI automations for finance teams on Lumera. Today we're opening up the platform to the finance community. If you're tinkering with AI for finance and accounting use cases, we want you in. Early access is live. Send this to your finance team! lumerahq.com/community
English
0
2
4
240
Ajaya
Ajaya@aj_agr·
@MarcJBrooker @felixhuttmann On a side note, I requested for a quota increase for agentcore - 177211276100066. Don't know if it's possible to get it bumped up before GA - but would be great to just build on top of the platform I can trust from the get-go.
English
0
0
0
42
Ajaya
Ajaya@aj_agr·
@MarcJBrooker @felixhuttmann If you allow custom images for code interpreter, rename it to sandbox - almost all the problems get solved I think. :) If you need vanilla sandbox, you have it - but we also have this cool runtime called agentcore which provides a bunch of primitives to build AI agents.
English
1
0
0
45