How AI Is Reshaping IT Services from the Inside
- 01The MSP Market Is a Decade Behind
- 02Services-First, Software-Second: A New Company-Building Archetype
- 03AI's Biggest Untapped Well Is Legacy Infrastructure Integration
AI + a16z | Joe Schmidt & Peter Doyle (CEO, Treeline)
1. Key Themes
The MSP Market Is a Decade Behind — And the Gap Is Widening
The managed service provider space, a $100B+ market, is paradoxically one of the most technologically underdeveloped categories despite being a technology category. This lag is structural, not accidental, rooted in decades of legacy process and tool accumulation. As AI accelerates, the gap between what's possible and what's deployed is growing faster than most realize.
"This managed service provider space, how IT and security services is delivered today is at least a decade behind where generally modern technology is. And with the pace of AI and just software very quickly getting better and better, that gap is only widening." — Peter Doyle [00:12:48.790]
Services-First, Software-Second: A New Company-Building Archetype
Rather than building a SaaS tool and selling it into the market, Treeline started as a services company on day one with zero software written. The thesis is that you must own the workflows and operational guts before you can meaningfully automate them — and that the transition from services to software should be iterative and compounding, not a big-bang deployment.
"Day one of Treeline was a services company, right? We hadn't built or written any software. But over time, if we do this right with well-built software, automation, AI agents, we can continue to pull what we're doing more and more towards being a software company." — Peter Doyle [00:09:45.670]
AI's Biggest Untapped Well Is Legacy Infrastructure Integration
The prevailing narrative around AI impact has focused on new apps and individual productivity. The far larger and largely untapped opportunity is integrating AI into the trillions of dollars of existing software infrastructure underpinning the global economy — and that requires new business models, not just better models.
"The biggest well of impact that I think is just starting to be tapped is there's trillions of dollars of software infrastructure that's been built over decades that underpins the U.S. and the global economy. AI impacting, integrating with, modernizing in any way, like interacting with this whole pool of critical systems and production environments is going to take a long time and new business models to do." — Peter Doyle [00:21:01.370]
2. Contrarian Perspectives
"SaaS Is Dead" — Or at Least Deeply Overapplied
Most people reflexively defend SaaS as the gold standard model. Peter argues that many categories — especially services-heavy ones — have too much SaaS, not too little. The proliferation of 30-35 stitched-together point solutions creates more operational debt than value.
"In a funny way, a lot of categories, most I would argue, and especially the one that we're in, has too much SaaS in a funny way, leverages too much software... those 35 tools aren't operating and firing on all cylinders. They're kind of like half-baked, half-configured, purchased 10 years ago." — Peter Doyle [00:05:12.890]
AI Rollups Are Not Venture-Scale Outcomes
The conventional wisdom in AI investing right now is that rolling up legacy services businesses and layering AI on top is a compelling venture strategy. Peter directly challenges this, arguing it's financial engineering dressed up as innovation — incrementally valuable but not category-defining.
"My hunch is no. I do think that to build a compounding business that is actually continuously innovating while compounding, you can't just rely on that acquisitive growth... it's a little bit more of an incremental approach to creating equity value versus actually creating some outsized opportunity to change the space." — Peter Doyle [00:24:08.170]
The Companies Most at Risk from AI Are Not Who You'd Expect
Common wisdom says AI threatens large, complex incumbents. Peter argues the opposite: companies with wide enterprise adoption and deep workflow entrenchment are safest. The most vulnerable are narrow, team-level or individual productivity SaaS tools — the ones people buy with a credit card and can just as easily cancel.
"If you're swiping a credit card to buy a few seats for your team or for yourself and it's something that's like not core to how your company as a whole operates, I think that's where you're going to be in trouble." — Peter Doyle [00:31:18.190]
Being a Services Company Is Not a Weakness — Even for AI-Native Businesses
Against the prevailing pressure to be "pure software," Peter argues that having humans in the loop is not a compromise but a structural feature. Even the frontier labs (Anthropic, OpenAI) are deploying hundreds of people to enterprise customers, validating the model.
"I was glad to see it because I really don't think businesses should be afraid to have people, services as a part of their offering to the market. Even if you have these leading frontier labs going into Goldman Sachs with a hundred people, they're recognizing that yes, their growth rate and their scale is unprecedented, but they're still realizing that they need to throw people at the problem." — Peter Doyle [00:29:56.270]
Human Adoption Inertia Is the Biggest Constraint on AI Progress — Not Model Capability
Most AI discourse centers on what the models can or can't do. Peter's view is that the binding constraint is human and organizational change management, and that this is chronically underweighted — especially by people inside Silicon Valley.
"Just how the world slowly moves incrementally to adopt this change is just going to be the biggest constraint. And that's just going to take way, way longer. And I do think that it's harder to sit in San Francisco and like being on Twitter and actually clock that in the right way. There's just like a natural constraint to humans changing." — Peter Doyle [00:32:15.930]
3. Companies Identified
Treeline Modern managed service provider reinventing the IT/security services model for SMBs by starting as a services company and iteratively building software and AI automation into operations. Mentioned as the subject company being built — positioned as a category-defining play in a $100B+ market.
"We're setting out to basically say, well, with modern technology, modern tools, and now AI, how can we essentially reinvent this category, not fully depart from the services model, but inject software and automation and AI that sits next to humans to fundamentally transform this market." — Peter Doyle [00:02:57.710]
Brex High-growth fintech company. Mentioned as prior employer of Treeline co-founder Hussein, establishing his operational credibility and network before co-founding Treeline.
"He came from originally a security company before going to Brex and he just was generally familiar with it." — Peter Doyle [00:12:01.590]
CrowdStrike Enterprise cybersecurity platform. Cited as a real-world example of the type of automated machine-generated security alerts that flow into MSP technician queues — illustrating the complexity and volume of signals MSPs must manage.
"It could be an automated machine generated alert from like CrowdStrike or some security tool that says you need to look into this vulnerability or this issue." — Peter Doyle [00:17:17.250]
ConnectWise Core PSA/RMM platform widely used by MSPs. Mentioned as the incumbent system-of-record that new entrants typically try to sit on top of — which Treeline explicitly chose not to do, instead going deeper into workflow ownership.
"Why can't you just go and offer a software platform to the MSP and just say, hey, we're going to make this all clean for you... We'll sit on top of your ConnectWise or whatever their core system is and make it better." — Joe Schmidt [00:06:59.490]
4. People Identified
Peter Doyle CEO and co-founder of Treeline. Former venture capitalist who left to build in the MSP/IT services space. Brings an unusual combination of investor pattern recognition and operator conviction to a market most Silicon Valley investors have ignored.
"From the venture capital seat, I was looking at a lot of software that sold into the space and there was, I think, one poor insight that really snowballed where everything else snowballed from." — Peter Doyle [00:12:19.750]
Hussein (Co-founder, Treeline) Co-founder of Treeline. Background in security (pre-Brex) and 15-year friendship with Peter Doyle. Described as the insider operator bringing deep IT/security domain expertise to complement Peter's investor perspective.
"We very much want to build Treeline by combining strong industry insider expertise with like new point of view and perspective... Hussein and I have been friends for 15 years." — Peter Doyle [00:14:13.190]
5. Operating Insights
Merge With Operators for Expertise, Not Revenue
When entering a complex services market, acquiring small legacy operators for their talent, domain expertise, and technician capacity — rather than their cash flows — is a faster path to building a credible offering than hiring from scratch. This is a distinct and underused M&A rationale.
"We wanted to merge. We did merge with a couple select traditional service providers, less to tap into their revenue and their cash flows and more to say, well, we need strong industry operators. We need experienced technicians because the technician team is very much a part of our offering." — Peter Doyle [00:14:14.350]
Audit Why Processes Exist Before Automating Them
When going into the guts of a services business to automate, the first question should be "why does this process exist?" — not "how do we automate this process?" Many workflows persist purely because of inertia, and automating broken processes just enshrines the inefficiency.
"You really have to look at the processes of how these companies operate... But then also question them originally. Why do you have these processes? Why do you do this? And oftentimes we find that the answer is, well, it's because we've been doing it for 15 years." — Peter Doyle [00:07:58.930]
Build for a 10-Year Roadmap That Doesn't Depend on AI Progress
Architecting an AI-native business should be done such that if AI progress stopped today, the company still has a decade of execution ahead. This forces intellectual honesty about whether the business model is real or just riding a technology wave.
"Yes, if AI progress stopped fully today, we would still have a 10 plus year roadmap to execute on." — Peter Doyle [00:23:22.350]
6. Overlooked Insights
The Data Moat Hiding Inside IT Services
Peter briefly mentions — almost in passing — that Treeline is accumulating a rich proprietary dataset of how businesses operate, how IT issues surface, and how customers interact with their infrastructure. This is never the headline of the conversation, but it may be the most durable long-term asset: a corpus of operational and environmental data across hundreds of SMBs that no pure-play SaaS vendor sitting outside the workflow could ever collect.
"We're now capturing so much good, like information and data around how we operate, how we interact with our customers. And we can do so much with that information over time and through building trust with the customers we serve." — Peter Doyle [00:26:49.310]
This data layer — built organically through service delivery — could eventually power predictive IT management, benchmarking products, or underwriting-style risk models for cyber insurance. None of the SaaS competitors selling into MSPs will have anything close to it.
The "Forward Deployed Engineer" Model Is an Admission That Changes the Investment Thesis
Peter makes an offhand observation that even the most advanced AI labs — Anthropic, OpenAI — are deploying large human teams to enterprise customers. He frames this as validation of the services-plus-software model. But the non-obvious implication is larger: if frontier AI companies with the best models still can't achieve enterprise penetration without heavy human services layers, then the total addressable market for AI-native services companies (not just software) is vastly larger than most investors are currently underwriting.
"Even if you have these leading frontier labs going into Goldman Sachs with a hundred people, they're recognizing that yes, their growth rate and their scale is unprecedented, but they're still realizing that they need to throw people at the problem." — Peter Doyle [00:29:56.270]
This reframes the entire "services are low margin, avoid them" bias in venture — the correct framing may be that services-led companies with compounding software layers are the right architecture for the AI integration era, and they are currently being systematically undervalued.