One month running a portfolio of SaaS products with AI agents: the honest report
A month ago I decided to stop building one product at a time and start running a whole portfolio of MVPs in parallel. The idea was simple: instead of betting everything on a single product, launch many small bets, use AI agents for the repetitive work, and let data tell me which ones deserve more attention.
Here is what actually happened.
The setup
I run Inithouse (https://inithouse.com), a one-person venture studio. Every product is an MVP built with AI-assisted tools, mostly React SPAs. The portfolio spans different niches: from a prediction platform (https://watchingagents.com) where you deploy agents to track questions about the future, to a personalized song generator (https://magicalsong.com), to card games, pet art, and developer tools.
The AI agents handle content distribution, SEO monitoring, data collection, and some operational tasks. I make the strategic calls. They execute.
Top 3 wins
-
Distribution beats product quality at this stage. The products that got the most traction were not the most polished ones. They were the ones where I nailed the distribution channel early. Google Ads as a validation tool (not an acquisition channel) turned out to be surprisingly useful for reading demand signals fast.
-
Content compounds. I started publishing across multiple platforms consistently. Not the same article everywhere, but unique angles per platform. After a month, some posts started driving organic traffic back to the products. Not huge numbers, but the curve is pointing up.
-
AI agents are genuinely good at the boring stuff. Monitoring search console data, checking indexation status, distributing content to multiple platforms, tracking what competitors are doing. These are tasks I would have skipped entirely as a solo founder. Now they just happen.
Top 3 failures
-
I launched too many products without a clear acquisition channel for each one. Some products are sitting there with nice landing pages and basically zero traffic. Having a product live is not the same as having a product in the market.
-
I underestimated how much context AI agents need. You cannot just say "go do SEO" and expect good results. Every product has different keywords, different competitors, different user intent. Building that context layer took way more time than I expected.
-
Some niches are brutally competitive and I did not do enough research upfront. I walked into a few spaces where established players dominate every keyword. The lesson: validate the distribution channel before building the product, not after.
What I learned about working with AI agents
The biggest insight is that governance matters more than capability. The agents can do a lot, but without clear rules about what they should and should not do autonomously, things go sideways. I spent a good chunk of the month building guardrails: approval workflows, review steps, and clear boundaries between autonomous execution and human decision-making.
The second insight is that agents are force multipliers for a specific type of work: structured, repeatable, data-driven. They are terrible at judgment calls, creative strategy, and anything that requires understanding why a customer would care. Those parts are still 100% human.
What is next
I am doubling down on the 3-4 products showing early traction signals. The rest stay live but get minimal attention until something changes. I am also investing more in content as a distribution channel because it is the one thing that compounds and does not require a bigger ad budget every month.
The goal is not to have the most products. It is to find the ones worth going deep on.
If you are running a similar multi-product approach or using AI agents for operations, I would love to hear what is working for you. What surprised me most this month is how much of the work is not building, it is deciding what deserves your attention.
The honest pattern I keep hearing from founders doing this: the agents are actually fine - the breakdown is in the human orchestration layer above them. You need to know what each agent's success criteria is, when to intervene, how to prioritize across the portfolio when three things need attention simultaneously.
Without a central system for that, you end up context-switching between 8 dashboards, losing the efficiency gain you were supposed to get.
The portfolio model works when you have one view that gives you the pulse of everything - revenue trend, status, what needs human judgment this week - without visiting each product individually. That single command layer is usually the real bottleneck, not the AI.
The 'honest report' framing is valuable - most AI agent posts are either breathless hype or dismissive. The portfolio-with-agents model is genuinely interesting because the failure modes are different from single-product operations.
What I'd want to know: what's the ops overhead per product look like, and did agents reduce it or just shift where the complexity sits? Because portfolio management has a coordination cost - knowing which product needs attention, what each one's health status is, which clients are at risk across the portfolio.
The risk is that agents handle the execution layer but you still need a dashboard layer: revenue per product, NRR trend, client status, decisions pending. Without that, the portfolio compounds complexity even as individual tasks get easier.
Building a Solopreneur OS right now - and this exact problem (multi-product coherence for solo founders) is one of the design constraints. CRM, revenue dashboard, and projects need to work across multiple active things simultaneously.
After a month, what's your biggest bottleneck - the agent execution layer or the oversight/visibility layer?
The honest report framing is exactly right - most AI agent posts skip the parts where it breaks or creates more overhead than it saves.
The thing I'd add: the biggest unlock with AI agents running multiple products isn't the task automation, it's the observability. When an agent is doing something, it needs a structured place to log what it did and why. Without that you end up with a black box that's hard to audit and even harder to hand off.
I've been building a Solopreneur OS in Notion with a Projects database that each product lives in - the idea being that whether a human or an agent is running the project, the status, decisions, and blockers are all in one place. Makes the honest month-end report you wrote much faster to produce.
What's your current approach to tracking what the agents actually did vs. what you planned they'd do? That gap tends to be where the real learnings are.
The honest report format is the right call for this. The AI agent framing makes it easy to highlight automation wins - less easy to talk about where the agents fell short.
What I'd be curious about: how do you track portfolio-level decisions across products? With 4-5 SaaS running semi-autonomously, the hard part isn't operations - it's knowing which product is getting your attention and capital, and whether those allocations are actually right.
I've been building a Notion OS for solopreneurs (projects, CRM, decisions, revenue, weekly review) because AI handles execution well; founders still need to handle strategy, context, and resource decisions. Those don't automate.
What's the biggest thing the AI agents still can't handle in your portfolio operations?