Journal

Insights from
mission control.

On AI agents, productized services, and the operating system that runs your company.

Featured
Automation Thesis

A 10-person company now runs on 2 humans and 12 agents

A 10-person company can now operate with 2 humans and 12 agents. This isn't a thought experiment, it's already happening. We break down the math, the roles, and the economics of the zero-headcount company.

Apollo Space Research··9 min
AI Operations

Why we built Apollo Space

We didn't set out to build an AI product. We set out to stop drowning. This is the story of how running a consultancy with 14 tools, 6 dashboards, and zero patience led us to build an operating system that replaced half our workflows overnight.

Ian Soares··9 min
All Articles
Product Thinking

The slow death of a marketer's voice

You publish one real piece a week and quietly translate it into ten, and each translation is a tiny chance to sound a little less like yourself. We built the OS because nothing on the market was guarding that.

Apollo Space Research··11 min
Engineering

The hidden tax of parallel agents is a migration diamond

Six agents writing to one schema conflict in the database, not the code, and CI dies at "multiple heads."

Apollo Space Research··10 min
Product Thinking

The day someone quits, your company forgets how it works

Onboarding isn't broken because training is bad. It's broken because your company can't remember, and we got tired of watching the answer walk out the door.

Ian Soares··13 min
Product Thinking

The first thing a new hire should do is read the company

A great onboarding doesn't hand you docs, it already knows who you are by the time you log in.

Ian Soares··11 min
Engineering

An orchestrator that can't survive its own crash isn't one

A crash that erases the orchestrator's reasoning loses the one thing you can't rebuild.

Apollo Space Research··11 min
Product Thinking

We gave our agents a Slack and a Telegram. Then they started talking to each other.

Agents stop being tools and start being coworkers when they share your channels.

Ian Soares··10 min
Automation Thesis

Promotions are dead. Trust budgets replace them.

You won't promote an agent; you'll widen its trust budget one verified task at a time, and the same ledger should govern your people.

Ian Soares··10 min
Product Thinking

The feedback inbox is three jobs collapsed into one tired head

Nobody owns the two jobs that actually decide your roadmap. We built the company OS partly to dissolve that hour, not to sell a smarter dashboard for it.

Ian Soares··12 min
Engineering

Put a deterministic gate in front of your smartest reviewer

The cheapest defect-catch is a dumb script that checks two merged branches still boot before any judgment.

Apollo Space Research··10 min
Engineering

Every authenticated page must live inside the shell

A page that "works" but drops the navigation around it is a regression even when every test is green.

Apollo Space Research··10 min
Automation Thesis

The job description is becoming a spec file

For an agent, a role becomes a versioned, testable spec, and that changes how you design every job, including the human ones.

Ian Soares··10 min
Product Thinking

The no is where the intelligence lives

Every calendar that only knows how to say yes is a funnel pointed at you. We're building the thing that can decline, because protecting time is the job no tool was ever shaped to do.

Ian Soares··11 min
Engineering

The first feature every money-spending agent needs is a brake

An agent that can spend has to halt itself before the next call when cost crosses a line.

Apollo Space Research··11 min
Product Thinking

The price that only changed in the email

The price changed in the call; the proposal still quotes the old number.

Ian Soares··10 min
Engineering

The graveyard of orphaned branches was the biggest source of lost work

A crashed agent doesn't lose what it pushed, it loses what only it could see.

Apollo Space Research··10 min
Product Thinking

The best hire you ever made wouldn't survive your own screen

We didn't build a faster resume filter. We built an OS that already knows the shape of your best people, so the shortlist comes from the team you already trust, not the words in a job post.

Ian Soares··12 min
Automation Thesis

Stop measuring output. Start measuring outcomes the company can’t forget.

An OS that remembers every decision and its result lets you grade the outcome, not the activity.

Ian Soares··10 min
Engineering

Sometimes the right number of agents is one

Under a fixed budget with clean context, a single agent beats a swarm, fan out only when domain isolation is a hard requirement.

Apollo Space Research··11 min
Product Thinking

Reminders that don't fire once and forget

A real reminder watches for the thing to actually happen, then nudges again.

Ian Soares··10 min
Engineering

The error message said "not configured." It was configured.

The most expensive bug isn't the failure, it's the failure that names the wrong cause.

Apollo Space Research··10 min
Automation Thesis

Who owns the work when the agent does it?

Autonomy without accountability is a liability; here's the org design that fixes it.

Ian Soares··11 min
Product Thinking

The standup is forty years of asking the wrong source

Status that's already written down should be read, not recited from memory. We didn't build a better standup, we built the company OS, and the morning blocker-chase is one of the first things it dissolves.

Ian Soares··11 min
Engineering

We resolved a three-way design argument by refusing to argue

Run all three specs against identical scenarios and let the numbers fold two of them.

Apollo Space Research··11 min
Product Thinking

Your quietest customer is your loudest warning

Churn announces itself in a slow fade of activity months before the cancel; a reactive CRM logs only the cancel.

Ian Soares··10 min
Engineering

"It’s a pre-existing failure" is how you ship your own regression

Red CI you didn’t cause hides red CI you did, unless you diff the exact failing set before and after.

Apollo Space Research··12 min
Product Thinking

A settings page is an unfinished decision

Every toggle you ship is a choice you couldn't make, handed to a user who can't either.

Apollo Space Research··4 min
Automation Thesis

Headcount was always a proxy

You never wanted 50 people; you wanted the work 50 people do.

Ian Soares··10 min
Product Thinking

The most honest thing a customer did all month became a tick on a chart

Feedback dies in the gap between a customer telling you something true and anything changing because of it. We didn't build a better survey. We built the loop that refuses to let that gap exist.

Ian Soares··12 min
Engineering

Trust is a ladder, not a switch

A new agent earns autonomy the way a new hire does, one verified task at a time.

Apollo Space Research··10 min
Product Thinking

The 9-minute window where every deal is won or lost

A human asleep, in a meeting, or off-hours cannot win a race a coworker never loses.

Ian Soares··10 min
Product Thinking

Dogfooding found the gap a thousand tests missed: we tried to hire into our own OS

You don't discover what's missing by testing the system, you discover it by living inside it.

Ian Soares··10 min
Automation Thesis

The founder's last real job is taste

When agents do the building and the selling, the one thing that doesn't delegate is judgment.

Ian Soares··11 min
Product Thinking

The cancel that never took: why your bill lies and nothing catches it

Money doesn't leak out of companies in dramatic mistakes. It leaks through the gap between what you were told and what you're actually paying, a gap nothing on the market was ever watching. We built the thing that watches it.

Apollo Space Research··12 min
Engineering

An eval is the only honest definition of "it works"

"It works" is a feeling; a flow that runs the real agent is a result.

Apollo Space Research··12 min
Product Thinking

The handoff is where work goes to die

Nothing breaks inside a lane; everything breaks in the gap between two lanes, a coworker lives in the gap.

Ian Soares··12 min
Automation Thesis

The subscription was the wrong unit

You weren't overpaying for seats; you were paying for the wrong thing.

Ian Soares··10 min
Engineering

It passed every test locally. It was broken the second it deployed.

Local-green is a fast proxy for the truth, never the truth, the deployed surface is the only judge.

Apollo Space Research··10 min
Automation Thesis

The real margin in AI isn’t the model. It’s the memory.

Anyone can rent the same model; the durable margin is the company brain you accumulate per customer.

Ian Soares··12 min
Product Thinking

The money you lose to dates nobody was watching

Auto-renewals, missed notice windows, lapsed deadlines, the recurring loss that has no owner. We built a system that watches the documents you stopped reading, because nothing on the market actually does.

Ian Soares··12 min
Engineering

Report two numbers or you're reporting none

A fast local eval is a proxy and the deployed surface is the truth, collapse them and you ship the proxy.

Apollo Space Research··10 min
Automation Thesis

The decision you made on Tuesday is already stale on Thursday

A reactive app never tells you the premise changed, a coworker re-checks the math.

Ian Soares··9 min
Automation Thesis

The company brain is the new system of record

The system of record stops being a database and becomes a memory.

Ian Soares··9 min
Engineering

The agent confirmed it set the reminder. The reminder never fired.

A scheduled job has two legs, it fires and it delivers, and confirming the first lies about the second.

Apollo Space Research··10 min
Automation Thesis

You'll buy your suppliers' agents before you buy their software

The next vendor relationship is their agent and your agent settling work between themselves, and procurement has no playbook for it.

Ian Soares··10 min
Product Thinking

The CRM was always going to rot, because you built it to be filled in by hand

A dirty CRM is a latency problem, not a discipline problem. So we stopped building forms and started building a system that records the truth at the instant it exists.

Ian Soares··11 min
Engineering

Your vector search is the quietest tenant leak in your stack

You can lock every table and still leak across orgs at the embedding layer.

Apollo Space Research··10 min
Product Thinking

Your agent should read your 3 a.m. Cloud bill

The spend that ruins a month shows up at 3 a.m., not in a dashboard you check.

Ian Soares··10 min
Automation Thesis

You don't need a cheaper chatbot. You need an OS that runs a company with eight people.

The leverage is an OS that does the operating, not a smarter tool that waits to be operated.

Ian Soares··10 min
Engineering

The backfill that ran inside a request and killed the app mid-demo

The slowest thing your agent does should never hold the connection the user is waiting on.

Apollo Space Research··11 min
Automation Thesis

Your next hire has no salary and no ceiling

The new unit of capacity is a role you switch on, and the only cost that scales is the work it actually does.

Ian Soares··10 min
Product Thinking

The Monday report is a question nobody asks

The grind we kept handing to our sharpest people was never the report. It was the silence after it, and that silence is what we set out to end.

Apollo Space Research··10 min
Engineering

Compaction is a decision you make before the window fills, not after

Prepare the summary in the background so the boundary is crossed without a stall.

Apollo Space Research··10 min
Product Thinking

The promise that died in the backlog

Most promises have no date, so if they only enter the backlog, they are gone.

Ian Soares··10 min
Automation Thesis

A 12-person company will get the AI-OS before the Fortune 500 does

A small company is the only org light enough to let one OS hold all of its work.

Ian Soares··10 min
Engineering

Two agents wrote to the same file. We almost shipped a database that wouldn’t boot.

Parallelism doesn’t fail loud, it fails in the one slot two writers both reached for.

Apollo Space Research··11 min
Automation Thesis

Five products, one engine

One engine, a different landing page per vertical, because mixing them kills conversion for all.

Ian Soares··10 min
Product Thinking

The clause you'll never see is the one that costs you

Reading every contract fresh, as if you'd never seen the last eleven, is the part that was always going to hurt you. We built a company OS that remembers, so the boilerplate stops stealing the attention the new clause needed.

Ian Soares··11 min
Engineering

Execution is local, coordination is networked

The safest agent control plane has no "run this" endpoint, by design.

Apollo Space Research··9 min
Product Thinking

Your tools each hold a third of the story. Nobody holds the whole one.

Tasks fall through cracks because no human and no app sees email, chat, and calendar at once, a coworker does.

Ian Soares··10 min
Automation Thesis

Everyone's building a governance layer for agents. They forgot to build the thing it governs.

Governance is a feature of an OS, not a product you buy before you have one.

Ian Soares··10 min
Engineering

"Is it working?" Has a different answer at every layer of the stack

A feature can be true in the database, true in the API, and a lie on the screen, answer per layer.

Ian Soares··10 min
Automation Thesis

A company is becoming a standing process, not a building of people

The native-service company is a process that runs continuously and surfaces to a human only when judgment is required.

Ian Soares··10 min
Product Thinking

The SDR role is a symptom, not a job

We didn't set out to build a better SDR. We set out to build the company OS, and the grind that burns reps out is one of the first things it dissolves.

Ian Soares··9 min
Engineering

The hardest reviewer on the team isn't human

The best contributor to a codebase is the one who says "not yet, you didn't test this."

Apollo Space Research··11 min
Automation Thesis

The most expensive email is the one nobody opened

A reactive inbox punishes you for not looking; a coworker reads the room while you sleep.

Ian Soares··10 min
Automation Thesis

The real race isn't a smarter model, it's who becomes the place work happens

Every model converges; the durable moat is being the surface where the work actually lives.

Ian Soares··10 min
Automation Thesis

The coordination tax is the only tax worth cutting

Most of what a growing company spends isn't on doing work, it's on keeping humans in sync; agents on one memory pay almost none of it.

Ian Soares··10 min
Engineering

We run a cross-tenant probe nobody asked for, because doors get found.

"Isolation is on" is a sentence; a probe that reads another tenant's data and fails is a defense.

Apollo Space Research··11 min
Product Thinking

The 4 a.m. Page is a measurement error

Nobody was supposed to be awake at 2:55 a.m. when the pool started creeping. We didn't build a smarter pager, we built the thing that stands in the hour no human can.

Apollo Space Research··13 min
Engineering

Never let a model grade its own homework

A judge model scores its own family higher, so a same-family eval measures loyalty, not quality.

Apollo Space Research··10 min
Automation Thesis

Proactive without permission is just a robocaller

The line between a coworker and spam isn't how often it speaks, it's whether it earned the context to speak at all.

Ian Soares··10 min
Automation Thesis

If agents are the new apps, something has to be the new OS

Apps need an OS underneath, and a model API is not an OS any more than a CPU is Windows.

Ian Soares··11 min
Engineering

Our long-running agent started confidently making up its own history

When you compress an agent's memory to keep it alive, the thing it forgets first is what it actually did.

Apollo Space Research··11 min
Automation Thesis

Accountability doesn’t move to the agent. It moves up.

Autonomy raises the altitude of responsibility, it doesn’t delete it.

Ian Soares··11 min
Product Thinking

The customer you lost on Tuesday was already gone in March

Churn isn't a surprise to the data. It's a surprise to the only thing in the building that couldn't watch, the human.

Ian Soares··12 min
Product Thinking

Undo is what lets you delegate

You hand an agent real work the moment a mistake costs a click, not a meeting.

Apollo Space Research··4 min
Engineering

RLS was on. The database still handed over the wrong tenant’s data.

Row-level security doesn't know who the customer is until your app tells it, the breach lives in the identity bridge, not the policy.

Apollo Space Research··11 min
Engineering

Your vector index is quietly going stale

Embeddings drift, content changes, the model updates, and the index keeps confidently answering with yesterday. Re-embedding is the maintenance job nobody put on the calendar.

Apollo Space Research··11 min
Product Thinking

Your status page should have written itself an hour ago

Incident comms are written last, by the most stressed person in the room, when they should be the first thing a system drafts the moment the signals diverge.

Apollo Space Research··10 min
Engineering

A prompt is a program you forgot to version

An unversioned prompt is an unversioned program: a one-word edit can silently regress a behavior nobody can see until a customer hits it. Treat prompts like the code they are.

Apollo Space Research··11 min
Automation Thesis

Your next customer will ask an LLM, not Google

Buyers now ask a model for the shortlist, and the model never sends them to your page. The citation is the new backlink, so write claims a model can lift verbatim, not keywords a crawler can rank.

Ian Soares··10 min
Engineering

Your eval set is a museum. It should be a drain.

A fixed test set rots the moment you start passing it. The fix is a flywheel: every real conversation that went wrong becomes a new test, so you can never overfit to a problem you already solved.

Apollo Space Research··11 min
Automation Thesis

Your docs are your new sales team

When a buyer asks an AI which tool to use, the model answers from your documentation, so your docs are no longer support content, they are the first sales call, and the reader who matters most can't ask a follow-up question.

Ian Soares··10 min
Engineering

Your agent should be able to see the screen

The last mile of 'do everything' is the software with no API. An agent that can see the screen and click reaches the work locked behind UIs that integrations will never touch.

Apollo Space Research··12 min
Engineering

When two agents disagree about a fact

When two agents hold conflicting truths in memory, you don't flip a coin, you run a reconciliation rule. Truth in a multi-agent system is a protocol, not an assumption.

Apollo Space Research··10 min
Automation Thesis

You'll hire agents the way you hire contractors

Headcount is a bet you make for years on a person you met for an hour. An agent you source for a scoped outcome, keep if it's good, and drop if it isn't, turns that bet into a much smaller one.

Ian Soares··11 min
Engineering

When one agent should become many

Fanning a task out to a swarm of sub-agents is not free, it buys parallelism and pays for it in coordination. The skill is knowing which side of that trade a task is on.

Apollo Space Research··10 min
Automation Thesis

When execution is free, judgment is the whole product

Once anyone can build anything overnight, the scarce good stops being the doing and becomes the deciding, what is worth doing and what good looks like. That is the only moat left to dig.

Ian Soares··11 min
Automation Thesis

When labor is software, the P&L stops looking human

Once the work is done by software, headcount leaves the income statement, payroll migrates into cost of goods sold, and gross margin starts to behave like a product's instead of a service's.

Ian Soares··11 min
Engineering

We let an agent plan its own week

Give an agent the right to schedule its own work and the dangerous part isn't the work, it's that nothing tells it to stop. The lesson is the brake, not the gas.

Apollo Space Research··13 min
Product Thinking

Voice is the interface proactive work was waiting for

Typing is a pull interface; you go to it. Proactivity is push; it comes to you. An OS that speaks first needs a channel you can answer without stopping, and that channel is your voice.

Ian Soares··11 min
Engineering

Three skeptics beat one judge

A single reviewer asked 'is this right?' nods along to the plausible-but-wrong. Three independent reviewers each told to prove it wrong, kill on a majority, catch what one judge waves through.

Apollo Space Research··10 min
Product Thinking

The tool in your stack that quietly went end-of-life

End-of-life is a date someone published, you just weren't subscribed to it. A watch on your stack's lifecycle surfaces the sunset while you still have room to move.

Apollo Space Research··11 min
Automation Thesis

The ten-person unicorn runs on specs, not headcount

When every person commands a fleet, the question stops being how many you can hire, and becomes how clearly you can say what good looks like.

Ian Soares··10 min
Product Thinking

The shared inbox where emails go to die

An email to support@ with no owner is a parcel addressed to nobody. Give every message one owner and one clock and the black hole becomes a thing that always gets answered.

Apollo Space Research··10 min
Automation Thesis

The seat was a unit of furniture, not a unit of work

Per-seat pricing is a fossil of the office floor, you paid for a place to sit because a place to sit was the only thing you could count. Agents do work without sitting, and the unit has to follow.

Ian Soares··10 min
Product Thinking

The report is late because the gathering is manual

The weekly report eats a Friday afternoon, not because the thinking is hard, but because the assembling is. Automate the gather, keep the judgment, and the report arrives early.

Apollo Space Research··10 min
Automation Thesis

The org chart is about to invert

When agents do the work, management stops being assignment-and-oversight and becomes taste-and-judgment, the pyramid that pushed orders down flips into a funnel that pulls quality up.

Ian Soares··11 min
Product Thinking

The step everyone skips because nobody is watching the checklist

A checklist only catches the thing you remember to check. The step that decides whether a customer stays is the one nobody watches, until skipping it stops being possible without a deliberate choice.

Apollo Space Research··10 min
Automation Thesis

The Berkshire of the agent era owns operations, not equities

The next great holding company won't buy businesses and leave them alone, it will run a studio of small companies on one shared brain and one operating system, where the moat is the OS, not the balance sheet.

Ian Soares··10 min
Engineering

The model is the cheapest part of the agent

Swap the model and your company keeps running. Swap the harness around it, the memory, the tools, the evals, the gate, and it stops. The moat was never the model.

Apollo Space Research··12 min
Product Thinking

The metric that drifted for a month before anyone looked

A number nobody watches daily moves a little every day until it is a crisis. A sentinel that knows what normal looks like flags the drift on day one, not in the quarterly review.

Apollo Space Research··11 min
Product Thinking

The meeting should end with the tasks already filed

The real cost of a meeting isn't the meeting, it's the homework afterward: the transcript nobody re-reads, the action items somebody re-types, the owners and dates that quietly never get set.

Ian Soares··10 min
Automation Thesis

When your team is agents, you stop assigning and start editing

When the team is agents, you stop assigning tasks and start reviewing drafts. The core skill flips from delegation-and-tracking to judgment-and-feedback, which is the work of an editor.

Ian Soares··11 min
Product Thinking

The context switch should be the machine's problem

Every interruption that lands on a person costs them the thread. An agent can take the hit and lose nothing.

Apollo Space Research··4 min
Automation Thesis

The last app you install is the one that installs the rest

Stop installing forty tools and wiring them together. Describe the job out loud and let the operating system assemble the capability, app-as-noun gives way to capability-as-verb.

Ian Soares··11 min
Engineering

The kanban is the agents' shared memory

When many agents work at once, the board stops being a human afterthought and becomes the one place they agree on what is true.

Apollo Space Research··11 min
Product Thinking

The inbox is a to-do list strangers wrote for you

Email inverts authorship: anyone can append a task to your day. Proactive triage gives authorship back, the system decides what earns your attention before you open the lid.

Ian Soares··10 min
Engineering

The heartbeat is the contract

A long-running agent doesn't prove it's alive by finishing. It proves it by leaving a trail you can resume from, the checkpoint it writes before it crashes is what lets the work survive the crash.

Apollo Space Research··11 min
Product Thinking

The follow-up is the job. The meeting was the easy part.

Booking the meeting is the part anyone can do. The chase that runs for days afterward, the reply that never came, the nudge nobody sent, is where deals actually die, and it's the part nobody owns.

Apollo Space Research··11 min
Product Thinking

The expense report nobody filed

The receipt landed in an inbox and never became an entry. Proactive capture makes the chore finish itself, done before you'd have remembered to start.

Apollo Space Research··10 min
Automation Thesis

When the buyer can run the eval, the demo theater ends

A demo proves a curated happy path on the vendor's data, on the vendor's schedule. An eval proves the real one on yours, in front of you. When the buyer can run the eval, the theater ends.

Ian Soares··11 min
Engineering

The deny-list never converges

You can't block your way to a safe agent, there is always one more phrasing. The only fence that holds is the list of things the agent is allowed to do.

Apollo Space Research··11 min
Product Thinking

The decision got made in a thread. Then it vanished.

The real decision happened in chat and now contradicts the doc, because companies capture decisions where they're written down, not where they're actually made.

Apollo Space Research··9 min
Product Thinking

The contract that auto-renewed at a worse rate

The danger was never the cancel that failed. It was the clause that quietly re-priced against you, and a watch on renewal terms, not just dates, catches the increase before it bites.

Apollo Space Research··11 min
Automation Thesis

The death of the dashboard

The dashboard was a pull interface for a push problem. You will not check it; it will tell you, the moment the number that matters actually moves.

Ian Soares··11 min
Automation Thesis

The company brain is a data network effect of one

Network effects usually need a crowd. A company brain compounds inside a single company, every interaction makes it sharper, and the moat is built entirely out of your own work.

Ian Soares··9 min
Automation Thesis

The browser tab is the last manual integration

Every time you alt-tab to copy a number from one tool into another, you are the integration layer, an API made of attention. An OS removes the human as glue.

Ian Soares··12 min
Product Thinking

The best proactive agent knows when to stay quiet

Proactivity isn't the art of speaking first, it's the art of suppressing the nine things not worth saying so the one that matters lands.

Ian Soares··11 min
Engineering

Most of your company is trapped in pixels

The work that matters lives in screenshots, scanned PDFs, and charts no API exposes, and reading those pixels is a different job from clicking them, harder and far more useful.

Apollo Space Research··11 min
Engineering

The agent that builds its own tools

When the agent writes the function it needs instead of waiting for someone to ship it, the line between using a tool and making one disappears, and the guardrail moves to where it belongs.

Apollo Space Research··11 min
Automation Thesis

The agency model breaks when the agency is software

Agencies sell the hour with a markup. When the work is done by an agent, the hour disappears, and the only agency that survives is the one that becomes a product.

Ian Soares··11 min
Product Thinking

The admin panel becomes a conversation

Settings pages, config screens, and dashboards exist because software couldn't understand what you wanted. When it can, the whole administrative surface collapses into a sentence.

Ian Soares··11 min
Automation Thesis

Software is about to stop being a noun

You bought tools, nouns, and assembled the work yourself. Capability is becoming a verb you ask for, and the app recedes behind the job done.

Ian Soares··11 min
Automation Thesis

The software budget and the payroll budget are becoming one line

IT spend and headcount spend lived on separate lines because a tool and a worker were separate things. An agent that does the work is both, and the moment those two budgets merge, procurement is never the same.

Ian Soares··12 min
Automation Thesis

Silence is not status: why 'no news' is the most dangerous update

A coworker reports the absence of activity that should have happened.

Ian Soares··10 min
Engineering

Route the cheap model first

Cost is not a setting you pick for your app. It is a decision you make per step, a small model to triage, a large one to judge, and most steps never reach the large one at all.

Apollo Space Research··12 min
Engineering

Retries are a story you tell yourself about reliability

A blind retry doesn't make an action reliable, it makes it happen twice. Real reliability is an idempotency key and a check that the effect landed, not re-running until no exception comes back.

Apollo Space Research··12 min
Engineering

RAG is not memory

Retrieval finds a document. Memory holds what the company learned. They are two different jobs and two different stores, and confusing them is why your agent feels amnesiac.

Apollo Space Research··11 min
Automation Thesis

Outsourcing was always about latency, not cost

We told ourselves we outsourced to save money. We actually outsourced to buy capacity we couldn't hire fast enough, and when the wait for capacity drops to zero, the whole logic of what you keep in-house flips.

Ian Soares··10 min
Engineering

Observability for agents is a different sport

Logs tell you what the code did. They cannot tell you why an agent decided to do it, and for a non-deterministic system, the why is the only debugging unit that matters.

Apollo Space Research··11 min
Engineering

Memory is not the context window

The window is RAM, fast, small, wiped between runs. Memory is disk, durable, curated, written on purpose. Stuffing the window full is not remembering; it is forgetting more expensively.

Apollo Space Research··11 min
Engineering

Llms.txt is the robots.txt of the agent era

Search crawlers obeyed a file at the root of your site for thirty years. The models reading your site for customers now have no such file, which means they read you wrong, and you never find out.

Apollo Space Research··11 min
Engineering

Latency is a product decision, not an engineering one

A deep agent that thinks for three minutes can feel like a coworker or a frozen tab, and the difference is never the speed. It's whether you told the user what the wait is for.

Apollo Space Research··10 min
Automation Thesis

Integrations were a confession

Every integration you wire up is an apology for software that couldn't talk to itself. An operating system makes the apology unnecessary, because the work already lives in one place.

Ian Soares··10 min
Automation Thesis

Instant is the new baseline

When the agent on the other side never sleeps, 'we'll get back to you' stops sounding polite and starts sounding broken, the customer's clock resets to now, and batched companies feel slow by comparison.

Ian Soares··10 min
Engineering

A senior engineer doesn't read the code. Neither should your agent.

An agent dropped into a codebase it's never seen wins by grepping for the seam, drawing the map, and reading three files, not by reading all of them.

Apollo Space Research··11 min
Engineering

Letting an agent run code is a containment problem, not a trust problem

Stop asking whether you trust the agent. The right question is what happens when it runs the wrong thing, and the answer is a sandbox that assumes it will.

Apollo Space Research··11 min
Engineering

One writer per working tree

Parallel agents that share a checkout don't crash loudly, they overwrite each other and lose work silently. The cheapest rule in our build kills the most expensive failure: a writer that gets its own isolated tree.

Apollo Space Research··10 min
Automation Thesis

The support tier was a triage system for a scarcity that's gone

Tier-1, tier-2, tier-3 existed because senior answers were rare. When the best answer is encoded and always-on, every customer gets a senior on the first contact.

Ian Soares··10 min
Engineering

Determinism is a feature: the boring parts are code, not prompts

Anything that has to be exact, auth, money math, routing, belongs in code; only judgment belongs in a prompt. A reliable agent is a thin language core inside a deterministic shell.

Apollo Space Research··11 min
Product Thinking

You've already answered their RFP. You just can't find it.

An RFP response isn't writing, it's retrieval and assembly from everything your company has already said, raced against a deadline. The agent drafts grounded in your own past answers, not blank prose.

Apollo Space Research··10 min
Use Cases

Can Apollo write your investor update?

Yes, because the hard part of the monthly update was never the writing. It was remembering what actually happened. Apollo reads the company and drafts; you keep the judgment and the tone.

Apollo Space Research··11 min
Product Thinking

The churn was decided three signals ago

An account doesn't die at renewal. It dies when usage drops, the tone sours, and the tickets spike, three weeks earlier, in three different tools nobody reads together.

Apollo Space Research··11 min
AI Operations

An agent earns the next job by showing the receipt

You do not hand an agent more responsibility because it is smart. You hand it more because you can see what it did.

Apollo Space Research··4 min
Engineering

The CVE in your build was public for months before it reached you

Dependency upgrades are the tech debt nobody owns, so they pile up until one becomes an incident. The fix is continuous: watch the advisory, open the PR, run the tests, surface the breaking change.

Apollo Space Research··11 min
Use Cases

Can Apollo triage your security alerts? The one real signal was buried in ten thousand

Tier-one security work is not catching attackers, it's drowning in alerts that aren't them. An agent that dedups, enriches, and suppresses the known noise hands you back the one signal a tired human missed.

Apollo Space Research··11 min
Product Thinking

Procurement is an approval process wearing a spreadsheet

Buying a thing for your company is four jobs of chasing and watching that nobody wants, and a human only needs to do one of them: say yes.

Apollo Space Research··13 min
Use Cases

Can Apollo run your partnerships desk? Yes, because BD is a memory problem

Business development is not high-volume outreach. It's research, a warm intro, a joint pipeline, and a nudge to the deal that quietly stalled, paced by the relationship, not the quota.

Apollo Space Research··11 min
Use Cases

Can Apollo run your customer onboarding?

Onboarding is a watched checklist with deadlines and stakes, the highest-churn window you have, and a proactive agent chases each step until the customer reaches first value.

Apollo Space Research··11 min
Use Cases

Can Apollo run your community without killing it?

A community dies the moment it can tell a bot is talking. The job isn't to answer everything, it's to triage the answerable, surface the unanswered, and hand the angry to a human before they feel handled.

Apollo Space Research··12 min
Use Cases

Can Apollo redline your contracts? Yes, against your playbook, clause by clause

Redlining isn't reading a contract. It's grading each clause against the deals you've already agreed to, flag the off-market term, draft the counter, name the risk, and handing a human the call.

Apollo Space Research··12 min
Product Thinking

The webinar ended at 11:42. The follow-up should already be out.

Event ROI doesn't die on stage, it dies in the follow-up nobody has energy left to do: segment the attendees, send each the right next step, open a task per hot lead.

Apollo Space Research··11 min
Product Thinking

The revenue that churns without anyone cancelling

A failed payment is a customer you keep losing every month, silently, with no decision and no email. Recovering them is the rare growth that costs nothing: smart retry, the right note at the right hour, and a chase that doesn't quit.

Apollo Space Research··11 min
Use Cases

Can Apollo process your refunds? Only the ones the policy already decided

A refund is not a judgment call, it is a policy with an audit trail. The agent settles the in-policy ones in seconds, routes the edge cases to a human, and logs every decision either way.

Apollo Space Research··11 min
Product Thinking

Your Portuguese is always three releases behind your English

Localization fails because it ships as a one-time pass and the source keeps moving, the fix is a loop that re-fires every translation the moment the English string it came from changes.

Apollo Space Research··12 min
Product Thinking

An audit is a binder somebody builds at midnight

A compliance audit is not a test of whether you're compliant, it's a deadline race to find the proof you already have. Apollo treats it as exactly that: map every control to its evidence, surface the gaps early, and let a human sign the page.

Apollo Space Research··11 min
Product Thinking

Your CRM doesn't need to be filled in. It needs a janitor.

Every CRM rots because data hygiene is a chore nobody owns, so give it to an agent that dedups, enriches, and backfills from the email trail, every day, forever.

Apollo Space Research··11 min
Product Thinking

Your docs rotted the moment you merged

Documentation goes stale at the speed of code. An agent that watches the diff and flags the now-wrong doc turns documentation from a debt you pay down into a thing that stays alive.

Apollo Space Research··11 min
AI Operations

The agent you can ignore

The goal isn't an impressive agent you watch. It's a boring one you forget.

Apollo Space Research··3 min
Product Thinking

A forecast you have to open is already late

Forecasting isn't a number on a dashboard you refresh. It's the alert the moment the curve breaks from its own pattern, with the reason attached.

Apollo Space Research··10 min
Use Cases

Can Apollo do your competitive intelligence?

Yes, because the job was never reading the competitor's page. It was noticing the one line that changed, on the morning it changed, instead of checking twelve tabs every Friday.

Apollo Space Research··10 min
Use Cases

Can Apollo close your books at month-end?

Month-end close is not arithmetic, it's a deadline-driven hunt for missing receipts, unmatched lines, and numbers that don't tie. The assembling and the chasing are agent work. The sign-off is yours.

Apollo Space Research··12 min
Use Cases

Can Apollo build your board deck? Yes, it's company memory, rendered

A board deck isn't a writing task. It's your company's memory pulled into one view: the metrics, the narrative, and the honest list of what changed since last quarter.

Ian Soares··10 min
Use Cases

Can Apollo be your IT helpdesk? Yes, because tier-1 was never a judgment call

Most help-desk tickets are the same five deterministic, auditable requests over and over, password resets, access grants, the FAQ. The agent should close those and hand the rest up with the context already gathered.

Apollo Space Research··11 min
Use Cases

Can Apollo be your QA department? Yes, because QA was always an eval loop

QA is not a phase you bolt on before launch, it is a loop: define the flow, run the real product, grade the result with a judge that never wrote the code, file the gap as a task.

Apollo Space Research··12 min
Use Cases

Can Apollo be your chief of staff?

The Chief of Staff job is mostly memory and follow-through, what slipped, who owes what, which decision went stale. That is not a person you hire. It is a process that never sleeps.

Ian Soares··12 min
Engineering

An agent that can’t forget can’t learn

Unbounded memory isn't intelligence, it's noise burying the signal. The agent that keeps everything forever ends up knowing nothing useful. Forgetting is the feature.

Apollo Space Research··11 min
Engineering

We do not let agents touch the internet by default

An agent that can reach anything can leak anything. Egress is a capability you grant per tool, not a default the agent inherits the moment it boots.

Apollo Space Research··11 min
Automation Thesis

What a company looks like when memory is free

Half your org chart is a workaround for the fact that human memory is scarce and lossy. When institutional memory is free and perfect, the shape changes.

Ian Soares··10 min
Engineering

A tool is a promise. Give it a schema.

A tool with a loose interface is a promise the model can break, a schema at the call layer turns that promise into a contract the system can enforce, so the model retries instead of hallucinating a shape.

Apollo Space Research··11 min
Automation Thesis

An operating system has a scheduler. Your AI tools don't.

The thing that makes Windows an OS and a calculator an app is a scheduler that runs work without being asked.

Ian Soares··10 min
Automation Thesis

The org chart was a caching strategy for slow humans

You drew reporting lines to route information through people who couldn't hold it all, agents on one brain don't need the boxes.

Ian Soares··11 min
Engineering

The worker said "done." A human clicked it. Nothing happened.

An agent's "STATUS: COMPLETE" is a claim to interrogate, never a fact to forward.

Apollo Space Research··10 min
Product Thinking

The fourth email you never send

Collections runs late even on perfect tooling because a person is standing between the overdue invoice and the send button, flinching. That flinch is the bottleneck, and it's the kind of thing only a company OS can take off a human's plate.

Ian Soares··12 min
Engineering

Your agent didn’t forget. It lied about remembering.

When an agent compacts its context, task-continuity survives but fact-continuity dies, and it fills the gap with a confident invention.

Apollo Space Research··11 min
Automation Thesis

Most outages happen with no alert at all. Your dashboard was never the problem.

Alerts tell you what already broke; a coworker notices the thing that has no alert yet.

Ian Soares··10 min
Automation Thesis

"Do everything" stopped being a red flag

A do-everything tool is a feature, not a red flag, the moment it integrates where you already are.

Ian Soares··10 min
Automation Thesis

Um sistema operacional para a sua empresa, não para o seu computador

Por quarenta anos o sistema operacional cuidou dos seus arquivos. O próximo cuida da sua operação.

Ian Soares··10 min
Product Thinking

The difference between a chatbot and a coworker is who speaks first

A chatbot waits to be asked. A coworker tells you the meeting moved before you walk into the wrong building.

Ian Soares··6 min
Automation Thesis

For 40 years your OS ran your files. The next one runs your operations.

For forty years the operating system managed your files. The next one runs your operations.

Ian Soares··9 min
Product Thinking

The 5-person company is the new 50-person company (and why that is a trap)

Tiny teams now ship enterprise surface area, and the failure mode flips from too-few-hands to no-one-who-can-explain-what-the-agents-did.

Ian Soares··8 min
Product Thinking

A notification is a tax. A coworker pays it for you.

Every alert is unfinished work pushed onto you, the software could not finish the job, so it interrupts you to finish it.

Ian Soares··9 min
Automation Thesis

Vertical AI is eating SaaS. The survivors will not be apps.

When agents replace point-tools one workflow at a time, the winner is not the best vertical agent, it is whoever owns the substrate every vertical agent shares.

Ian Soares··9 min
Product Thinking

Can Apollo turn a sales call into a spec and a pull request?

The transcript already contains the requirement; the rest, the task, the spec, the branch, the draft PR, is a pipeline, not a meeting.

Apollo Space Research··9 min
Engineering

The day our agent filed its own bug

The agent hit a broken tool, apologized to the user, opened the ticket itself, and by morning the fix was merged and it simply finished the job.

Apollo Space Research··8 min
Automation Thesis

Your first agent should embarrass you

A one-tool agent running in production this week beats a brilliant six-month plan that touches nothing.

Apollo Space Research··4 min
Product Thinking

Can Apollo run your content engine? (We used it to plan this blog)

This post exists because a radar agent mined the topic, a writer agent drafted the piece, and a reviewer agent tried to kill it, the content pipeline is the first employee we hired from our own product.

Ian Soares··9 min
Product Thinking

Donna from Suits is a product spec

The fictional assistant who knows everything before you ask is not fantasy, she is a one-message morning briefing: the three emails that matter, the meeting you cannot miss and why, the date about to bite.

Ian Soares··10 min
Automation Thesis

Reactive software was a 40-year detour

Software learned to wait, for a click, a query, a prompt. The next platform does not wait, because work does not.

Ian Soares··8 min
Engineering

Your agents have a credential-sprawl problem

Bolting a dozen tool servers onto a chatbot does not give you an AI company, it gives you a dozen credential leaks and a context tax; an OS owns the connections so the agent does not have to.

Apollo Space Research··9 min
Engineering

We build Apollo with a council of agents

Several agents write code. Two refuse to let bad work merge. One keeps them all honest. The hardest reviewer on the team is not human.

Apollo Space Research··9 min
Automation Thesis

The native service company is the real AI business model

Don't sell a seat and don't sell hours, sell a productized service that runs itself, with humans only where judgment earns its keep.

Ian Soares··9 min
Automation Thesis

The system of record is becoming the system of action

For twenty years software stored what already happened; the next layer does what happens next, and when the layer that acts outranks the layer that records, the CRM becomes a database the OS reads, not a product anyone opens.

Ian Soares··8 min
Automation Thesis

Most AI pilots die in production. The model was never the problem.

Pilots fail structurally, not technically: a brilliant agent with no operating system underneath it is a smart intern with no desk, no badge, and no memory of yesterday.

Ian Soares··9 min
Automation Thesis

Spawn your kind-of-Hermes-agent, Apollo's thesis

The gap between 'AI is a tool' and 'AI is a coworker' is the AI-OS wedge. Here is what we are building at Apollo to close it, and why we think the rest of the market is still building the wrong half.

Apollo Space Research··11 min
Engineering

A bigger context window is not a memory

Everyone is chasing bigger models. The real competitive advantage is memory. After six months, your Apollo Space SDR agent knows your ICP better than any new hire. That knowledge compounds. Models don't.

Apollo Space Research··15 min
Engineering

One agent is an engineering problem. Twelve is a coordination problem.

Running one AI agent is an engineering problem. Running twelve is a coordination problem. When your SDR agent and your competitor watch agent disagree about timing, who wins? Here's how we solved it.

Apollo Space Research··15 min
Automation Thesis

SaaS pricing is dead, agents charge per outcome

Per-seat pricing made sense when software was a tool. But agents don't sit in seats, they deliver results. The shift from 'pay for access' to 'pay for outcomes' is the most important business model change since SaaS itself.

Ian Soares··10 min
Engineering

Your monitoring was built for software that waits

Your monitoring stack was built for services that respond to requests. Agents don't respond, they decide, act, and learn. Here's why your observability needs a fundamental rethink.

Ian Soares··13 min
Engineering

Anatomy of an AI agent: memory, tools, and decision loops

Most people think an AI agent is an LLM with a system prompt. It's not. An agent is memory, tools, and a decision loop. Here's how we built twelve of them.

Ian Soares··14 min
AI Operations

What 800 meetings taught our team intelligence agent

Apollo Space's Team Intelligence agent has processed over 800 meetings. The patterns it found, overstaffing signals hidden in silence, churn warnings buried in tone shifts, cash burn encoded in meeting frequency, changed how we think about management.

Apollo Space Research··11 min
Use Cases

Teaching an AI agent your brand voice: lessons from 6 months of content

We spent six months teaching Apollo Space's content agent to write like us. It started terrible. The internal read rate went from 12% to 94%. Here's every mistake we made and every lesson we learned.

Apollo Space Research··14 min
AI Operations

How our meeting digest agent killed standups, and nobody missed them

We had 4 standups a week. Then our Meeting Digest agent started auto-summarizing every call, extracting action items, and updating the CRM. Within a month, attendance dropped to zero. Here's what happened next.

Apollo Space Research··10 min
Use Cases

How our competitor watch agent saved a $200K deal

A deal was dying. Our competitor watch agent flagged a pricing change no one was tracking. The sales team used the intel, repositioned the pitch, and closed $200K in ARR. Here's the full timeline.

Apollo Space Research··12 min
AI Operations

Twelve agents, one founder: running a company without a team

What does it actually look like to run a company with AI agents instead of a team? This is a day-in-the-life account, from the 6 AM dashboard check to the midnight agent report. No theory. Just what happened.

Ian Soares··11 min
Use Cases

The QA agent that catches what humans miss: a PR-by-PR breakdown

Automated tests check if your code works. QA agents check if your product works. Here are five real bugs Apollo Space's QA agent caught that passed human review, CI pipelines, and unit tests.

Apollo Space Research··14 min
AI Operations

Dashboards show you what's broken. Agents fix it.

For 20 years, SaaS has been building better windows into your problems. Dashboards show you what's broken. Agents fix it. The era of passive software is ending, and most companies don't realize it yet.

Ian Soares··11 min
Use Cases

Our SDR agent booked 47 meetings in a month. Here's what it learned.

We gave an AI agent our CRM, our email, and our LinkedIn. Month one: 12 meetings. Month three: 47. This is the data-driven story of what changed, what broke, and what surprised us.

Apollo Space Research··11 min
Product Thinking

Most "AI agents" are chatbots with a database

Most products labeled 'AI agents' are chatbots with a database. Real agents don't wait for prompts, they observe, decide, and act. Here's why this distinction matters more than any other in the AI landscape.

Ian Soares··10 min
Product Thinking

Taste over prompts: the real skill behind great AI products

Everyone is learning prompt engineering. But the real differentiator isn't how you talk to the model, it's knowing what to build and what to leave out. The gap between 'technically works' and 'delights users' is taste.

Ian Soares··13 min
Product Thinking

The risk isn't an agent that fails. It's one that succeeds at the wrong thing.

The biggest fear with AI agents isn't that they'll fail. It's that they'll succeed at the wrong thing. The answer isn't restricting them, it's building trust architecture that lets them earn autonomy the way new employees do.

Ian Soares··13 min
Product Thinking

The hidden cost of building your own AI agents

Every CTO asks the same question: should we build our own AI agents or buy a platform? The answer is nuanced, but the hidden costs of building are not. Here's a practical framework based on what we've seen work and fail.

Apollo Space Research··12 min
Product Thinking

Why your AI pilot failed -- and what to do instead

80% of AI pilots fail to reach production. Not because the technology doesn't work, but because companies treat AI like a feature instead of a workflow redesign. Here are the five anti-patterns killing your AI initiatives.

Ian Soares··12 min
Automation Thesis

The agentic internet: when software talks to software

The internet was built for humans. The next internet is built for agents. When your SDR agent negotiates with a client's procurement agent, we'll need new protocols, new economics, and new trust models.

Apollo Space Research··13 min
Automation Thesis

The death of the integration: why agents don't need APIs

We spent two decades building APIs, webhooks, and integration platforms to make software talk to software. Agents made all of it optional. They don't integrate with tools, they use them.

Apollo Space Research··13 min
Automation Thesis

The Latin America AI advantage: why emerging markets lead agent adoption

The conventional wisdom says Silicon Valley leads AI. The data tells a different story. Companies in Latin America are adopting AI agents faster than their US counterparts, not despite their constraints, but because of them.

Ian Soares··12 min
Automation Thesis

Software ate labor. Agents ate software.

Marc Andreessen said software is eating the world. He was right. But he missed the next act: agents are eating software. The SaaS era created 30,000+ tools. The agent era collapses them into one orchestration layer.

Apollo Space Research··12 min