Elliot

Partner at Harbor Software Group. Prev founded @agnostiqhq - high performance compute orchestration for AI/ML (acq. by DataRobot).

[Elliot @ElliotMacgowan]

@toddsaunders I think the near term practical instantiation of this is a “headless mode” for the VMS coupled with an SE that would really allow customers to drive the implementation themselves - with supervision of the SE - and without requiring tons of custom code

[Elliot @ElliotMacgowan]

@Mid0 @geoffintech Cool thanks!

[Elliot @ElliotMacgowan]

@JaredSleeper Lots of agent observability companies still in the embryonic stage but certainly coming

[Elliot @ElliotMacgowan]

@NextBigTeng Good take. Im curious what insulates “agent SaaS” from the same effects? To me ai-native SaaS apps are just an intermediary step towards new kinds of software that we aren’t even contemplating yet.

[Elliot @ElliotMacgowan]

@dharmesh Agreed. There are just so many more interesting things to build now, I dont understand why you would waste your precious attention vibe coding your own SOR. I also feel like folks forget the cost structure for tokens - GPUs - which are not exactly free.

[Elliot @ElliotMacgowan]

@claudeai Man this is sick. Been using a homegrown (way worse) version of this. Have to say the UX is almost exactly what I envisioned. Soooo long Anthropic

[Elliot @ElliotMacgowan]

@geoffreylitt Been toying with something similar. It feels like the most natural UX for broad ranging agentic tasks beyond just coding

[Elliot @ElliotMacgowan]

@businessbarista @Austen Eat more of the end-user workflow and grow the TAM beyond the SOR. Ie eat more into labor than traditional SaaS.

[Elliot @ElliotMacgowan]

@levelsio Ive had the same realization and Google can thank @OfficialLoganK

[Elliot @ElliotMacgowan]

😳

[SemiAnalysis @SemiAnalysis_]

BREAKING CUDA MOAT EXPANDS: Today, NVIDIA has acquired SchedMD, makers of SLURM, a widely used "open source" workload scheduler. Many AI companies such as Mistral, Thinking Machines, parts of Meta's FAIR division, university academic labs use SLURM. NVIDIA's acquisition expands the CUDA moat to a new layer, as many customers trying to use alternative AI accelerator chips like AMD, Intel, etc. also rely on SLURM (a lot use k8s too). Unlike Run:AI & DGX Lepton, which nobody uses, a lot of people use SLURM. (1\2) 🧵

[Elliot @ElliotMacgowan]

@lessin @andruyeung 💀

[Elliot @ElliotMacgowan]

@l3amm Congrats!

[Elliot @ElliotMacgowan]

@DBredvick Can you give us a sense of the steps the SDR agent worked through? Did they touch email personalization? If so, which features did you feel the need to “hardcode” most to fit your business vs a vendor solution

[Elliot @ElliotMacgowan]

@yacineMTB Just stay awake is actually top tier parenting advice

[Elliot @ElliotMacgowan]

@TheStalwart Replit

[Elliot @ElliotMacgowan]

@grebtso @GavinSBaker OpenAI and others will trace the “chain of thought” data that gets created when using a reasoning model - i.e, the part when the model speaks to itself, and use it as additional training data.

[Elliot @ElliotMacgowan]

Can someone intro me to @harleyf pls?

[Harley Finkelstein @harleyf]

Can someone intro me to @SorayaMartinezF pls?

[Elliot @ElliotMacgowan]

I am a former AI infra founder (a la $CRWV) turned VMS HoldCo founder (a la $CSU). AMA.

[Elliot @ElliotMacgowan]

@mikealfred @PhadsEth He’s making a subtler point about the steps between building the shell and turning the clusters online. I.e. the work typically done by OEMs and SIs. Hes giving credit to both the DC builders and the cloud operators.