Chris Menge

- AI Tender Suite
- Search Everywhere Optimization

[Chris Menge @cmenge]

@Neffton1 Sei froh. Kann auch eine Ersatzzustellung an einen Notar werden der Dir nochmal für 30s Googlen eine Rechnung über 660 netto plus Auslagen plus Mwst. stellt, plus Gerichtsgebühren…

[Chris Menge @cmenge]

@om_patel5 Yesterday, I had it try a service via MCP that it had built before. It told me it was using it "in anger" and then rushed to fix the bugs it had built because working with the MCP interface was so annoying it seems:

[Chris Menge @cmenge]

@weaverobotics Lovely song! No worries, I won't shut the door and replace the keys.
When will you come back, Dave?

[Chris Menge @cmenge]

Toying around with an investment tool. Still prototyping, it's currently a sidecar to Claude Code.

Anybody got any experience with "soft" FinTech, how to market etc.?

The cool thing is that I use this myself a lot, so this is the ideal case for dogfooding... Only reason I didn't buy the other positions yet is timing (Japan already closed).

Not financial advice - do your own research.

[Chris Menge @cmenge]

Are there concrete names for the institutional investors?

I have a few concerns:
- SEK 700M directed issue, 12.28M new shares at SEK 57, priced today (upsized from the SEK 600M intention yesterday), ~9.7% discount to the June-30 close, Pareto accelerated bookbuild?!
- It's the third raise in ~2.5 months (April placement → SEK 125M @ SEK 14.50 in May → SEK 700M now).
- Proceeds fund capacity for an "opportunity pipeline," not booked orders. Management sold equity at ~2.2x April / ~4x the May placement into the hype because the business can't self-fund — and the stock closed today (~SEK 55.9) below its own SEK 57 placement price.
- Insider dump: head of wireless sold his entire 1.39M-share stake (~SEK 99.5M) on May 29; three directors resigned.

What about the Ningi report, and the the various probes? [Swedish EBM + Finansinspektionen (listing-leak); US class-action firms (Rosen, Bronstein)]

I'm probably seeing this in a very negative light, but these are too many yellow-to-orange flags for me.

[Chris Menge @cmenge]

@TeLaicite Airport Bus, Vienna… one degree below “rare” for veal…

[Chris Menge @cmenge]

Fable is super impressive. First model that let me casually max out my Max 200 subscription's session limit, and it did it within maybe 25 minutes...

[Chris Menge @cmenge]

@audiencon https://t.co/IZnWIAvCBX

[Chris Menge @cmenge]

@om_patel5 To be fair, Opus is an expensive model and Anthropic generally relies on heavy context caching. The first message (including claude.md, mcp etc.) is heavy and has no caching; subsequent requests benefit from 90% discounts (at least if they follow the APIs pricing)

[Chris Menge @cmenge]

@CRSegerie IQ or not, 18 months ago, I was happy when it wrote some working code that wasn't completely weird. Today, I'm annoyed when, after 20 minutes of work that would have taken me a week, it has a few bugs left and it didn't cross-check some changes in 8 other parts of the app...

[Chris Menge @cmenge]

Temperature / electroweak leaves a 'trace', namely photons -> classical behavior. Cold fullerenes leave no trace -> interference.
So what about gravity? Could we do a gravitational two-slit interferometry experiment with particles much larger than C60 to find evidence for the (non-)quantization of the gravitational field? That's what MAQRO is trying to do if I understand correctly?

[Chris Menge @cmenge]

It's actually quite simple: if the path _could_ be observed, if behaves like a particle. If not, waves.

Proven with C60 molecules (fullerenes): the hotter they are, the more particle-like they behave based on their probability to emit an infrared photon on the path.

In other words, if you leave a trail, you're a particle. If you're not leaving a trail, you don't.
No time-travel or simulation hypothesis required.

Read up on decoherence and Zeilinger's work.

[Chris Menge @cmenge]

"True Multi-Agent Collaboration Doesn't Work"
Sounded like a dumb article to me, but the article and the paper linked were so good I wrote a full piece on it https://t.co/WaHVRykyhj

[Chris Menge @cmenge]

The numbers are brutal.

Same task. Same AI model. Same budget. Four agent org structures:

Single agent: 28/28 ✓
Hierarchical: 18/28
Concurrent: 9/28
Pipeline: 0/28

The most complex architecture spent its entire budget *planning* without writing one line of code 😵‍💫

[Chris Menge @cmenge]

@itsolelehmann Agree. The platforms are already closing their APIs, only a matter of time they'll crack down on RPA.
And overloading people with junk won't work anyway.

[Chris Menge @cmenge]

@csaba_kissi Judging from my own process, it's very close already.

What was absolutely impossible w/ AI coding 6 months ago and needed a lot of handholding is now a sidenote of an even larger task.

[Chris Menge @cmenge]

@tdinh_me How do people even do that? RPA? I mean the API blocks it, so you have to remote-control a browser, all for comments that alone don't have much value?

[Chris Menge @cmenge]

@chrija Contrarian take: most of the systems of record don't need any AI features. They do need a good agent-API such that a second layer of agents can interact with them: the SoR has hard business rules, the agent plane has the context and the instructional logic / goal.

[Chris Menge @cmenge]

@hamzafarukuguz Why even wait for a prompt? The agent can just watch Jira for new tickets and act whenever it sees something.