here i’ll share some mental models, tools and concepts to hopefully help you navigate this space

no time to read? click here and listen to a podcast episode i generated about this instead 🎙️

let’s kick off with autonomous businesses

a year ago no-one was talking about agents.
today no-one is talking about autonomous businesses (almost).
we’re not there yet, but we are getting there..

autonomous businesses are businesses where a.i. systems handle core operations and decision-making, with humans either serving in an oversight role or, in some cases, minimal involvement.

the winners of tomorrow will push to levels 3, 4, and 5.
here's what the journey looks like:

→ level 1: manual

everything runs on human effort. decisions, execution, and even record-keeping rely on people. the tech is minimal, if it exists at all.

→ level 2: assisted

basic automations start to emerge. repetitive tasks like data entry or scheduling get delegated to software like r1, chatgpt, perplexity, and claude, but humans still make all the big calls.

→ level 3: semi-autonomous

systems take over most day-to-day tasks.

• ad campaigns are launched + optimized automatically (yes this is possible and happening).

• customer service is managed through ai (see klarna or sunnycars).

• internal workflows (like approvals, task assignments, and updates) happen without human involvement using platforms like make, mindstudio, taskade, or more complex solutions like crew ai or n8n.

• lots of exciting stuff also happening on the compute-use front (ai operator and twin.so)
  

  humans step in only for high-level strat or manage edge cases. this is where businesses start moving faster than their human-only competitors.

→ level 4: fully autonomous (we’re not there yet, but we’ll get there)

systems don’t just execute, they make decisions.

• marketing campaigns are planned, launched, and adjusted without human input.

• inventory is restocked dynamically, forecasting demand based on trends and user behavior.

• customer retention strategies, pricing, and even new product recos are automated.

humans monitor and refine the system, but the machine runs the show. most of the business operates independently, humans focus on oversight, not ops.

the tools i mentioned above can already achieve some of this (if plugged correctl). also have a look at what’s happening with autogen (self feedback loops, semi-autonomous, can operate in teams) and langgraph (multi-step reasoning, memory, tool calling... no real autonomy but it's starting to smell like fire)

level 4: this requires engineering, self-evolving systems need custom solutions and continuous api management. the tools mentioned above can help.

→ level 5: self-evolving (we’re not there yet, but we’ll get there)

this is the pinnacle. systems learn and improve themselves continuously.

• ad campaigns optimize for creative and targeting in real time, based on live user feedback.

• supply chains refine themselves, adapting to traffic, weather, and market changes.

• customer experiences become hyper-personalized as systems automatically understand and adapt to individual preferences.

• entire org becomes a living system, capable of optimization without needing human intervention.

this is where businesses achieve constant improvement, not just automation.

we’re not there yet, i truly believe we will, and it takes getting through levels, 1, 2, 3, 4 to reach 5…

some businesses are testing level 5 in niche areas (especially saas and ecommerce), but very few are truly there. give it a year.

to recap:

at level 1, scaling is limited by human effort

at level 3, humans step back from execution

at level 5, businesses become living, adaptive systems that never stop improving

what tools? if you’re currently trying to get to level 2, 3, 4, 5?

here’s my current take (based on the input from amazing people like tim cakir, walid boulanouar, luísa lima, ross stevenson, adrian pradzioch and many many more)

so here's my quick list and current assessment....
tried to mental model it into:
easy to hard / 🐣 basic to 🦅omg...

note: it isn't about the list or which tool, it's that you start getting yours hands dirty (it will be dirty) with these

here it is with the links (and easier to read, i really need to change fonts)
agents... by simple to hard, simple to awesome

🐣
llms (r1, claude, perplexity, mistral, gpt-4 o1)…co-intelligence, raw reasoning, no orchestration
custom llms (customgpt, chipp etc = more control but single-instance ai)

🐥
make (triggers llms, not real agents, still super useful)
n8n (open source, flexible but still deterministic)
relevance ai (embeddings, retrieval, light logic, we're starting to get there)

🐓
flowise (no-code llm chaining, prototyping ai workflows fast, open source lacks deep agentic logic)
langgraph (multi-step reasoning, memory, tool calling... no real autonomy but it's starting to smell like fire)
haystack agents (good for rag = retrieval)

🦅
autogen (self feedback loops, semi-autonomous, can operate in teams)
langgraph (multi-step reasoning, memory, tool calling... no real autonomy but it's starting to smell like fire)
crew ai (multi-agent with roles and tasks, starts replacing humans)

🔮 bonus
autogpt (early attempts at full autonomy, not ready yet)
twin . so (tbd but claims to be real agentic infrastructure)

ok this is to automate decision-making and all the operations involved with running a business
what about building?

here’s my take

building with ai

here’s the breakdown with links
🐣
gamma (intuitive design tools, quickly create visuals and layouts, beginner-friendly)
relume (streamline web design, focuses on UI/UX components, simple integration)
framer (prototype tool, interactive designs, bridges the gap between graphics and code)

🐥
bolt.new (instant app creation, e-commerce focused, minimal coding required)
replit (online ide, supports many programming languages, good for learning and small projects)
lovable (focus on user-friendly designs, likely app development or website builders)

🐓
cursor (data operations simplified, less coding more management, streamlines complex tasks)
vercel v0 (deployment platform, integrates with development tools, supports next-gen web technologies)
windsurf (likely advanced data handling or AI applications, more code-intensive)

🦅
claude (advanced reasoning, potentially involves AI, higher complexity)
github copilot (ai-powered code assistant, enhances coding productivity, substantial coding knowledge needed)

🔮 bonus
placeholder for future advanced tools (future predictions or cutting-edge technologies not yet mainstream)

what’s crazy here in my opinion is this…

speed is the new edge

(even more than before)

eric schmidt says it best: speed is the new edge. it's so true.

in this video i tried to illustrate his point a bit. this tech is giving us wings 🪽

check out the video here

→ prototyped in a couple hour
what used to require endless briefings, dev + ux teams now takes one person a few hours (and yes it’s deployed)...i know it's not perfect... but it's done (perfect is the enemy of done)

→ landing page/website up in an afternoon
no need for a full-stack team, ship a page that converts, test it, iterate. what used to take weeks now takes hours. you can build out the entire website (about us, product, mission etc)

→ user onboarding flows/guides built in half an hour (6 of them)
automate everything. guide users without human support.

→ video assets generated in half an hour
content creation isn’t a bottleneck anymore. build great content, recycle at will. ai generates, repurposes, and adapts instantly. what used to be a slow, expensive process now requires some good taste (eval) and fantastic prompting (have you tried deepresearch?)

→ customer interviews... pre-ran them with synthetic interviews... this allows me to interact with a digital twin, i'm so much more prepared for when i have real customer interviews

here's a sentence from my synthetic persona:

"i need a straightforward ai playbook that shows me exactly what to automate, you know, which tools to use, and how to integrate them without needing an it degree or too much time invested. If I could set up ai workflows that improve efficiency without making marketing feel robotic, that would be a ideal, yeah i'd pay for that""

shared a bunch more in these 2 playbooks first one here, second one here

and this is just a sample…

i could also have recorded:
outreach campaigns, tested and automated
workflow automations that cut ops time by 90%
asset creation across formats
content marketing
website optimisation
app optimisation
feature ideation (driven by real-time demand signals)
customer support (handled 24/7, humans for edge cases)
retention hooks
sales automation
synthetic customer insights, feeding product + marketing
and more.............…

he future won’t be dominated by giant tech firms or big scale-ups. it’ll be millions of small, hyper-specialized startups, each just a few people, running lean, scaling fast.... here's how and why

the industrial revolution created massive orgs with 10k-200k employees. scale required size. overhead was a necessary evil.

but software changed that. and ai is about to finish the job.

in the future (and for some, already today), the average company will be 4-5 people.

→ a solo founder, a few key operators, and an army of ai agents running everything else
→ hyper-specialized teams, loosely connected through APIs, collaborating when needed
→ no bloated org charts. no endless meetings. just speed, execution, and results

venture capital will have to adapt. funding models built for 100-person startups won’t work, the ideal team is 5-10 people.

the winners? the builders who see it first.

→ 100,000s of people will generate wealth through 100,000s of agent-enabled ventures
→ scalable, profitable businesses with low overhead
→ ideas shipped without full-time teams, politics, or bottlenecks

the future isn’t a few big winners. it’s millions of small, efficient, automated companies.

this is already happening.

why? because software creation is becoming free.

→ for decades, building software was expensive.
→ developers were costly because they translated english into javascript.
→ now, llms are driving software costs to zero.

this changes everything.

lower costs = more startups = more niche solutions. one-size-fits-all models of massive corps will collapse under the weight of ultra-targeted products w/ specific engaged audiences.

the most successful founders? distributors first: masters of attention, audience, and growth. distribution.

the new startup model

→ a 5-10 person team (20 max max)
→ an ai-powered workforce running ops, marketing, sales, and support
→ lightweight, api-connected biz scaling with near-zero overhead
→ infinite experimentation: launching and iterating fast

so what happens to venture capital?

1. adapt to smaller, hyper-profitable, ai-drivn startups

2. double down on deeptech, robotics, energy…anything capital-intensive

the first path: micro-funds, revenue-based financing, and new funding structures built for 5-10 person startups running on ai.

the secnd path: traditions vc dollars flow into hardware, biotech & ai infras: where capital still makes a difference

the classic “$20m series a for 100-person saas startup”? disappearing

future of venture won’t be about funding headcount but $ automation, distributio & defensibility

this is already happening

teams are staying leaner. ai-native startups skipping the classic "hire fast, scale headcount" playbook

or do you still need to raise loads for distribution?

here’s where i start to wonder

→ if ai can replace large part of markting, sales, & ops, do startups even need $20m to scale?
→ if a solo founder can build a $10m ARR company with ai agents, does vc even make sense?
→ if automation drives growth instead of headcount, what exactly are vcs funding? marketing?

capital no longer the advantage? distribution, defensibility, and automation are. require as much capital?

this means the new funding game looks different:

→ micro-funds for ai-powered teams of 5-10 people
→ revenue-based financing replacing traditnl equity rounds
→ venture capital shifting toward hardware, deeptech, and ai infrastructure = places where cash is still king

the old model: spend spend to hire fast and dominate a market seems built for a different era.

possible twist:

capital flows into ai-native rollups → if startups stay small, we might see more holding companies, daos, or networked ownership models where multiple ai-powered businesses are loosely connected under a shared infrastructure

on 1 side: ai-powered, ultra-lean startups that don’t need massive funding or headcount to scale.

on the other: trillion-dollar giants that absorb everything. openai + microsoft + softbank merging to stay? ai arms race is already pushing companies & governments into deeper alliances.

>megacorps own the foundational ai models, chips, and infrastructure.
> micro startups build on top,
> governments struggle to regulate any of it

funding headcount is out. funding leverage is in.

and what moats are left in the age of ai?

quite a few actually

→ exclusive data. proprietary datasets no one else can access or replicate. if your ai is learning from something unique, you have an edge.

→ physical resources. ai still relies on real-world constraints. compute, lithium for batteries, satellite lanes.

→ regulation. love it or hate it, regulatory approval creates moats.

→ partnerships. it’s about better distribution. deep partnerships get access to exclusive data, resources, and customers.

→ supply chain control. owning critical infrastructure….chips, robotics, or logistic…creates leverage that pure software companies don’t have.

→ accountability. some industries demand responsibility. legal, insurance, and defense need traceability and trust. ai that serves these spaces will have built-in barriers to entry.

→ network effects: platform gets smarter as more people use it and competitors can’t easily replicate that growth, you have a real moat.

→ switching costs. think enterprise software, cybersecurity, or healthcare, become impossible to rip out once integrated.

→ brand and reputation. in a world flooded with ai-generated content, trust is the differentiator. a strong brand compounds over time.

→ distribution. owning the right channels, whether it’s retail, enterprise contracts, or api integrations, still wins.

→ rlhf (reinforcement learning with human feedback). tuning ai models with proprietary feedback loops makes them more useful over time. best models evolve with unique, high-quality reinforcement.

the strongest ai companies won’t just be the ones with the best models. they’ll be the ones with the best moats.

ai is changing the game. but moats still win it.

that’s it for today!!

hope you enjoyed this.

keep building keep tinkering

⬛️

david

ps: ❤️ enjoyed this email?

pls recommend the newsletter to a friend or colleague

and here are some previous pieces you might have missed