akshay hangloo

Head of Product Engineering @neatlogs.

I write about why agents break in production - traces, evals, cost, and the feedback loop nobody owns.

[akshay hangloo @akshay_hangloo]

Most teams open agent traces when a customer complains.

That's backwards.

Technical failures are easy to catch. A tool call errors → you see it → you fix it.

What actually kills agents in production is behaviour.

The agent takes a worse trajectory. Quality drifts. It hits a tool-call limit and quietly degrades instead of stopping.

No error. Just worse work.

The fix isn't better prompts.

Treat agent runs like a data pipeline:

→ Structured traces, captured by default → Auto-tagging that makes each pass more useful → Deterministic + non-deterministic logs together → Evals continuously running on real production runs → Human feedback steering AI evaluators

The goal isn't just observability.

It's making every run teach the system something.
There's another problem almost nobody solves:
The person best qualified to judge whether an agent did good work is often a domain expert who can't read a trace.

Today, their feedback ends up in a Slack thread.
A week later, it's gone.
That feedback needs to become structured data the system can actually learn from.

I've spent a decade in data engineering, then started building agentic systems where non-technical users found weak points within days.

Agents are a data problem wearing an AI costume.

Building the tooling for this at @neatlogs.

[akshay hangloo @akshay_hangloo]

@itsakaashhh Interesting marketing angle - CFO for AI spends :)

[athrv.sh @cultist_dev]

the worst bug in my agent failed steps that worked.

@neatlogs raised an incident on it: 25 "Execution Failed" detections against a baseline of 0.5. that is 52.8x, across 10 traces.

the investigation named the cause at 66% confidence. the planner wrote a step that ends in `cd my-project`, then checked for `my-project/mkdocs.yml`. the check ran from inside the new folder, so it looked one level too deep.

commit 2de852f checks files from the folder the step started in. the MkDocs quickstart went from a false fail to verified.

an agent that tests docs can fail a doc that is right. a spike alert on your own agent's failures catches that before a maintainer does.

#neatHack

@bettersayAJ @akshay_hangloo @EntireHQ @HaimantikaM @arithecfo @agent_wrapper

[Rithvik @BngRithvik]

😋would be mentoring for the @neatlogs
Neathacks hackathon
Register ASAP
Thanks to @simranrambles and @neatlogs

[akshay hangloo @akshay_hangloo]

@kurokotetsuyami run this on neatroots, & share your findings with me

[akshay hangloo @akshay_hangloo]

@reachjalil time to build jalilcrm in one afternoon :)

[akshay hangloo @akshay_hangloo]

Lazy ass observability tools are here to die soon!

[ajay yadav @bettersayAJ]

no observability tools were harmed in the making of this video

[neatlogs @neatlogs]

BLR on-site day 1

[akshay hangloo @akshay_hangloo]

Already shipped a couple of features on day 2 of neatlogs offsite. Way to go buddy!

[Rajdeep @r9jdp]

I'm now building at @neatlogs

Joined as an MTS intern with @bettersayAJ and team

big stuff coming up

[ajay yadav @bettersayAJ]

.@kartikeyjandial is hosting his first 12 angry devs. SF please be kind 😅

[akshay hangloo @akshay_hangloo]

@thepanshi_ @cengines_studio once in a blue moon, but good enough to create reels other than bad roads & traffic

[akshay hangloo @akshay_hangloo]

@thevibefounder @neatlogs super excited for this one!

[akshay hangloo @akshay_hangloo]

we are here to keep your agents working, not just running.

[ajay yadav @bettersayAJ]

great websites are still hard to build. we shipped a new one for @neatlogs today

[Kartikey Jandial @kartikeyjandial]

First week progress🐐🐐🐐

[neatlogs @neatlogs]

unfortunately we're the breaker of bad news. but hey it comes with a plan to fix :)

[ajay yadav @bettersayAJ]

here's a sneak peak of our pre-launch party.

@blader revealed @arithecfo and shared how the latest advances in AI enabled him to expand the magic behind runway to a much larger audience.

then i showed how @neatlogs keeps Ari reliable in production.

full video dropping soon...

[Antaripa Saha @doesdatmaksense]

doing my debut in video content with my fav topic -> evals

we are starting this series and will go a lot deeper into how to do evals. excited ;)

[neatlogs @neatlogs]

ep 1: how to read traces?

[ajay yadav @bettersayAJ]

threw a party for the IIT Bombay folks in SF, we thought it would be fun to get everyone together

it was even more fun than we expected. really good crowd, lots of old connections renewed and new ones were made, and nobody seemed particularly interested in leaving 😂

big thanks to @sahildhingra20 and @Sanskar3074 for hosting it with us and making the night what it was!

[simran sachdeva @simranrambles]

introducing semantic detections on @neatlogs.

it’s running on jev by @typesafeai

going live on neatlogs this week ! we’re rolling it out to a select group of customers first. comment below if you’d like access, and we’ll enable it for your account

[Simran Sachdeva @simranblethers]

my X account got hacked (insert a sad emoji with a cowboy hat)

please tag @X @Support

help me get my account back (@simranrambles) back