Patrick Mathieson 🇺🇸

An anxious moment for software open.substack.com/pub/thedownrou…

@BoxLongs I agree with the general sentiment but valuing Toast on the same EV/RR metrics as the others is deeply unserious

SAAS Claude + Guggenheim Top 5 "BEST" set ups. $MNDY $TOST $BRZE $FRSH $TEAM, Group Context The Guggenheim software universe NTM EV/Rec Rev median sits around 3.5-4.0x as of March 20, 2026, compressed substantially from the 6x+ levels of early 2025 and the 10x+ peaks of 2021. Average NTM revenue growth across the group is approximately 12-14%. The five names identified above all deliver above-average growth at below-median valuation, which is the core thesis: the selloff has created pockets where the market is discounting durable growth businesses at multiples historically associated with low-growth or declining revenue profiles. The two Buy-rated names (MNDY, TEAM) carry additional conviction from Guggenheim's fundamental coverage.

@ganeumann no it's just beginning

Well, it seems the making friends and influencing the powerful phase of my life is over

@DKThomp @pkedrosky I would have liked a little more humility about the fact that markets are difficult to predict, often mysterious, and tempting for humans to fit post hoc narratives to. Yes, since Jan MSFT is down 20%. But from April-Dec it was up 50%. There’s MANY factors behind those moves.

@DKThomp @pkedrosky I had a hard time with this one. Paul seems to view all stock movements in his favored directions as vindications of his views, and all stock movements the other way as an example of perverse market incentives. A lot of these claims are unfalsifiable.

New pod: 'YES, AI IS A BUBBLE. THERE IS NO QUESTION.' In the last few weeks, I'd swung from thinking AI is a bubble to thinking it's not. I wanted to gut check my change of heart. So I brought on @pkedrosky to make the 2026 bubble case. Absolutely packed episode. Paul somehow covers all of the following in an hour: - The meaning of the Mag7 reversal (the stocks that dominated returns in '24 and '25 are mostly down in '26) - The omen of the Saas-pocalypse - The private credit crisis, explained - Why the inference-revenue boom from new agents like Claude Code might be a sugar high (explosive rev growth now, much slower rev growth after SWE adoption peaks) - If value is leaving software, where is it flowing? (energy!) - "The reason productivity is rising has nothing to do with AI" - Plus, we debate: model improvements, the agent economy, the durability of chips, and what the bubble will look like, if it comes

@marcepntoja there have been barely any IPOs so the high growth / high multiple portion of the market is increasingly less represented in publics

Software is at 4.8x EV/NTM revenue. That is below the 2018–2019 average of 7.4x, when most of these companies were unprofitable and AI was a conference buzzword. The fashionable explanation is that AI agents will do everyone's job, and if there are no employees there are no seats to license. Perhaps. But it is also true that the pandemic pulled forward years of software adoption, juiced revenue growth to 30%+, and inflated multiples to 16x. Growth has since decelerated. Interest rates quadrupled. The most boring possible explanation is that the bubble deflated. The AI narrative is convenient because it gives the multiples compression a story. Valuation risk coming home to roost after a three-year melt-up is less interesting to write about. Both are probably true.

Cardinal Cupich @CardinalBCupich: "A real war with real death and real suffering being treated like it’s a video game — it’s sickening."

When the flag covered coffins start coming home (and they will), and these people try and do the usual thing and be solemn and pretend we had no choice but to do this, that this was a serious decision not lightly taken, etc. I think people will remember this shit

Ro Khanna: " I have never seen a more dishonest campaign than Donald Trump and JD Vance running as the pro-peace candidates."

“People are becoming more aware of the amount of money that ends up in the pockets of merchants of death. With that money, hospitals and schools could be built. Instead, those already built are being destroyed.”

The taking of human life is of the utmost seriousness. EVEN when the taking is justified. EVEN when the target is the enemy. Especially from leadership. Especially when killing at scale, or from a distance. This video is inhuman.

@battl2heaven no OL?

My post combine horizontal board for pick 18. Any questions, ask away. DL Peter Woods CB Jermod McCoy RB Jeremiyah Love DE TJ Parker DL Christen Miller WR Makai Lemon WR Omar Cooper CB Avieon Terrell FS Dillon Thieneman FS Emmanuel McNeil-Warren CB Mansoor Delane

Vikings will be picking 18th. Some initial names I like right now: IDL Caleb Banks IDL Peter Woods Edge Keldric Faulk ILB Sonny Styles CB Jermod McCoy WR Makai Lemon Edge TJ Parker Edge Cashius Howell Defense heavy just based on the players and the board.

@convequity partners are using it too. So the value comes from distribution and network density. As most founders know just because you’ve built something good doesn’t mean anyone else is going to use it. So the value of these systems comes from something different than cost+plus of the code

@convequity I think what’s missed here is the difference between single player and multi player software. The cost of software for one company or one individual to use is collapsing to zero. But the utility of sw with network effects is the fact that all my employees / customers /

The pace of depreciation for Software is accelerating. We discussed this in Oct 2025, long before the market action. But we discussed the theoretical cost of generating 10m lines of code is $3k. And it can be off by 100x or more depending on the AI progress & capabilities. Recently, Anthropic gives us very good data points on the work on Building a C compiler with a team of parallel Claudes. To build a functioning C compiler it takes: - 16 Parallel Agents (Docker) & 2B Input, 140M Output Tokens - 2,000 Sessions & ~2 Weeks - $20k API Costs - 100k Lines of Rust Clean-room (No Internet) And only Opus 4.6 can make it actually work: - EARLY OPUS 4: Barely working compiler - OPUS 4.5: Passed test suites, failed real projects - OPUS 4.6: Compiled Linux kernel So that means the actual cost per 100k lines of code for a fair complex foundational software component is $20k. And if we apply simply extrapolation, then 1m lines of code costs $200k. Obviously oversimplified as most models can't handle this large code base or will see cost skyrocket. And so far, only Opus 4.6 has the engineering capability to work on such a complex project. But this maturation of engineering capability is highly certain vs. uncertain progress towards AGI. We know that all it takes is just additional RL, inference then training loop, and ever larger clusters. So current AI training CAPEX is also likely not a bubble as well. Assuming we don't make any progress on making more efficient algo, and chips & infra will get 4x cheaper in the next few years, and models will be capable of working on more lines of code in the future, then we can truly see more software companies getting lower valuations. If a SaaS company has 20M lines of code, then in the incoming years, assuming 4x efficiency gain, the future Opus model can generate one functioning SaaS at $500k. That's about the annual wage for 2-3 developers, and it is quite a bargain especially for enterprise customers. convequity.com/notes-when-ai-…

this is so cool

@evantindell you're forgetting MNDY, would be at the top of this list

Recent software buyback announcements per Gemini... is $CRM gonna take on debt to get this one done? Its like 3x annual FCF

right. anything that can be accomplished in 5min of prompting will cease (on average) to be economically valuable

@villi we've seen it work

I often see founders thinking they can sell into private equity firms through which they can get access to their portfolio companies. It never works.

@Konstantine I would guess you need to be 1) enterprise, 2) multi-product, and 3) with a shared data architecture powering all your various products (e.g. how DDOG does it) so your product suite is really a platform

How will SaaS businesses evolve in the era of AI?, by @pmthn @thedownround/note/p-189023278?utm_source=notes-share-action&r=1cd1i" target="_blank" rel="nofollow noopener">substack.com/@thedownround/…

@yongfook this is reminding me so much of "blockchain solves this"

"...but in the future the agents will do that" is becoming the low IQ, deus-ex-machina response of someone who consumes way too much X content.