Pau Sabria

Co-founder of @RemotelyWorksHQ. Previously co-founder @Olapic (acq by Monotype). Alum @bcg @columbia

[Pau Sabria @kuap]

it's been 48h since i started using @lovable

now i'm the proud owner of 4 non-functioning web apps.

[Pau Sabria @kuap]

When our developers stopped getting raises, most stayed anyway.

Here's why.

The average contractor in our network stays 18 to 24 months, and often longer.

Yearly raises across LATAM have been thinning for years now, or they're much smaller than they used to be. That reflects what the market is doing to engineering pay, and we happen to be able to read it inside our own network.

Every retention model says this is how you lose people.

We didn't.

I think a raise and the certainty of being paid fairly are two different products. We confuse them because they usually arrive together.

What makes a good developer start looking is usually the slow accumulation of not knowing.

Are they still at market? Was the person hired six months after them brought in higher?

Nobody asks that out loud, and it compounds quietly for a long time before anyone updates a profile.

So at @remotelyworkshq we rebenchmark compensation against LATAM rates every quarter, and we show the developer where they land. In a flat year, they still know the number is honest.

Most of what gets called a retention problem is an information problem.

A raise might buy a year of goodwill. Knowing what you're worth buys the years after it.

[Pau Sabria @kuap]

Asking who's using AI is measuring the wrong thing.

Rolling out AI across a mixed team at @remotelyworkshq has shown me four kinds of people.

The gap between them is much bigger than the gap between using AI and not using it.

𝗧𝗵𝗲 𝗯𝗿𝘂𝘁𝗲 𝗳𝗼𝗿𝗰𝗲𝗿 knows what they want but not how to get there. They throw giant, multi-step prompts at the model, then blame the tool when the output falls apart. They panic the moment the AI takes initiative.

𝗧𝗵𝗲 𝗲𝗻𝗱𝗹𝗲𝘀𝘀 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻𝗮𝗹𝗶𝘀𝘁 spends hours talking to Claude. Nothing gets documented. Nothing gets saved. Every new chat starts from scratch after the context window closes. It feels productive, but produces little.

𝗧𝗵𝗲 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲𝗱 𝗹𝗲𝗮𝗿𝗻𝗲𝗿 has no technical background but knows to ask for a CSV with the right fields. Years inside tools like Clay and HubSpot taught them to think in systems. They improve quickly.

𝗧𝗵𝗲 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝗯𝘂𝗶𝗹𝗱𝗲𝗿 thinks in edge cases. They create their own workflows without being asked and are usually three steps ahead of everyone else.

The first two use AI constantly.
The last two get the best results.

AI adoption isn't the metric that matters.

The real difference is how people think.

That's why one rollout strategy fails for all four.

[Pau Sabria @kuap]

We've analyzed the commit histories of 25 million developers.

The actual code they wrote, across GitHub and GitLab, going back years.

Most staffing firms promise you the "top 1%." Measured how? A resume screen and a 45-minute interview. That tells you almost nothing about how someone actually works.

So at @remotelyworkshq we built Gitsight to read the thing that does. Across 25 million engineers, it looks for three signals no interview can show you:

Are they writing the hard parts of a system, or just the easy edges?

Do other engineers build on their code, or quietly route around it?

Is their contribution steady over years, or does it spike the month before they start job hunting?

None of that fits on a resume. All of it is in the history.

It took years to build. The data only exists if you started collecting it before you needed it. A resume is a marketing document. Years of work other engineers depended on are much harder to fake.

Most companies are still hiring on a performance. We hire on the record.

[Pau Sabria @kuap]

@diervo @MarkVillacampa 🍿

[Pau Sabria @kuap]

In the next 18 months, OpenAI will acquire Sierra and Brett will succeed Sam.

[Pau Sabria @kuap]

Founders, stop leading with equity.

We're seeing only 31% of companies in our network offer equity, down from 35% last year.

For new hires, nearly three out of four receive none at all.

It's now something companies reserve for tenured and leadership-track engineers.

So what actually gets a senior engineer to say yes?

✓ A competitive base salary.
✓ Guaranteed equipment.
✓ A smooth hiring experience.

Those are the parts of the offer that matter most.

The problem?

You can't adjust any of them if someone else controls compensation.

That's how most staffing models work. An agency sets the rate, adds its markup, and hands you a finished offer.

You're renting a seat.

At @remotelyworkshq our transparent cost-plus model puts you back in control. You set the compensation, shape the offer, and build the relationship directly with the engineer.

Lead with the parts of the offer you can actually control.

There are a few more surprises in the data.

Grab the full 2026 LATAM Software Engineering Salary Guide: https://t.co/peSbALRVZJ

[Pau Sabria @kuap]

@MarkVillacampa @diervo Bueno, y aun asi, Bret fue Co-CEO de Salesforce... it doesn't get any closer than that.

[Pau Sabria @kuap]

This chart from David Sacks is going around as proof that AI isn't killing jobs.

The data is right. Companies adopting AI the hardest are hiring more, including at the entry level. At @remotelyworkshq we placed more engineers in the first half of this year than in any comparable period.

But after six years of watching engineering teams up close, what concerns me isn't the hiring. Hiring is fine. What broke is the apprenticeship.

Seniors got made by the messy work: taking a first pass at something they didn't fully understand, breaking it in production, owning the incident, getting torn apart in review. That cycle, repeated hundreds of times, is what built judgment. And that work is exactly what AI now handles.

So you can hire a full class of juniors and, five years later, not have produced a single senior. They're sitting one layer above the work that used to teach them. The headcount looks healthy, but the learning loop is gone.

We interview hundreds of candidates every quarter, and AI is making this harder to catch. It inflates resumes, polishes interviews, produces clean coding exercises. We've seen a real increase in people who can perform "senior" for exactly as long as the conversation lasts.

AI can fake a clean codebase. It can't fake scar tissue -- the calls someone made when it mattered, the systems they kept running, the trust they earned over years. That's what we try to hire on, and honestly, it keeps getting harder.

Curious whether other founders and engineering leaders are seeing the same thing. How are you adapting?

h/t David Sacks, original chart below

[Pau Sabria @kuap]

The "senior developer" your agency bills you $70 an hour for is taking home a third of that, and already looking for the door. You won't hear it from the agency.

You'll get a vague email next quarter saying he's "no longer available." Then a replacement who needs two weeks to onboard and another two before he's actually useful.

The wider the gap between what you pay and what the developer earns, the more the agency makes. So the incentive runs backwards. Finding you the strongest engineer barely factors in. Finding the cheapest one they can bill at a premium does.

We've watched a developer billed at triple what they actually earned. When that developer walked, the client got two lines and no handoff. Just "no longer available," and a month of ramp-up to get back to where they were.

Your outcome was never part of the equation.

That model works beautifully. For them.

We built @remotelyworksHQ the other way around. Developers set their own rates and keep 100% of what they charge. We make money too. Just on a fixed fee that doesn't grow when the developer gets underpaid.

No quiet markup or opaque math.

There's nothing clever about it. It's how this should have worked from the start.

If you want developers who stay and do their best work, stop running a model that depends on underpaying them.

[Pau Sabria @kuap]

I've watched this happen more times than I can count.

A candidate comes through a trusted intro and suddenly your process gets 𝘢 𝘭𝘪𝘵𝘵𝘭𝘦 more flexible.

A reference call gets pushed back. You overlook a profile inconsistency. That identity check feels unnecessary because someone credible already vouched for them.

Yet a cold applicant usually goes through every step of the process. Profile review, technical assessment, references, identity checks, and whatever additional controls are needed for the role.

I've witnessed versions of this in hiring teams for years.

What I've come to believe is that verification isn't one decision. It's asking:

-Can this person do the job?
-Are they who they say they are?
-Is the same person who interviewed actually going to show up and do the work?

A referral can help answer one of those questions. It doesn't answer all of them.

That distinction matters even more now as proxy interviews, identity fraud, and candidate substitution become increasingly common in remote hiring.

The companies that handle this best separate trust, identity, and capability rather than treating them as interchangeable signals.

Fraud follows the path with the fewest checks.

@RemotelyWorksHQ, we've built our verification process around the assumption that identity, skill, and trust are separate claims.

Each needs its own proof, which is why every candidate follows the same minimum verification baseline regardless of how they entered the funnel.

[Pau Sabria @kuap]

Restaurant marketplaces work. Hairdresser ones don't.

I've spent years building a marketplace for senior engineers, and this is the simplest lens i know for why a few work and most quietly die.

In a marketplace, you want people coming back to search again and again. OpenTable works because you eat somewhere new on Friday and somewhere different next week. The habit continues on.

Hairdressers are the opposite. You try a few, you find one you like, and you stop searching. The few that survive, like Booksy and Fresha, had to add scheduling and payments on top. Pure discovery never worked.

Founders assume churn is the enemy. For a marketplace, repeat demand is the whole business model.

Software engineering sits in an interesting middle.

The average developer changes roles every 18 months to two years. That used to keep me up at night. It's turnover, and it looks like leakage.

The turnover is the point. The developer who leaves every 18 months is exactly what brings the client back. They always have a new seat to fill, a new team to staff. Recurring need is the raw material.

But recurring need isn't a moat by itself.

A client who liked one of our developers can go hire the next one directly. What keeps them with us? We refill the seat faster than they could on their own. A vetted bench, a replacement ready in days, no sourcing tax.

The thing that looks like it should kill the business only feeds it if you keep earning the next placement.

[Pau Sabria @kuap]

@XDeGracia @a16z Congrats on the raise! I wouldnt be surprised if you are back on the news a few months from now!

[Pau Sabria @kuap]

@caixabank necessito un pdf de titularitat del meu compte en vosaltres. Com teniu la falta de vergonya de cobrar 2,41eur i tardar 24h en generar-lo?!

[Pau Sabria @kuap]

The software development talent layer isn’t going anywhere either

[Aaron Levie @levie]

The app layer couldn’t get a better advertisement than a company spending $500M to build their own version of it. Obviously lots of nuance here that can’t be captured in the headline, but this should make you very bullish on software.

[Pau Sabria @kuap]

This should be issue number 1 for every political party in Spain. But it requires something every party lacks: thoughtfulness and pragmatism.

Instead, they hide their incompetence in plain sight by bringing forward a battery of significantly less important issues. The system is perverse because the more they can hide their incompetence, the longer they stay in power.

People will continue to have diminishing purchasing power, and politicians will deflect blame on others: immigration, foreign investors, corporations, etc.

[El Equidistante @elequidistante]

Esta mañana he encontrado una nómina vieja de mi padre revisando papeles de casa.

Marzo de 1992, ingeniero jovencito con 6 años de experiencia. Casado, con dos hijos e hipoteca en Madrid.

Por curiosidad me he puesto a hacer cálculos, y me ha dado permiso para compartirlos.

El bruto del mes eran 615.704 pesetas. Ajustando a IPC, hoy serían 120.000 € brutos al año equivalentes. Un ingeniero con ese mismo perfil cobra ahora entre 35.000 y 45.000 €.

Un tercio. Un puto tercio del sueldo real que tenía mi padre con su edad.

Pero donde la trampa se ve más clara es en la fiscalidad.

Mi padre, sumando IRPF y Seguridad Social, soportaba una carga fiscal efectiva del 27% sobre su bruto (24% IRPF + 2,7% SS, porque cotizaba al tope máximo). Le quedaban netos el equivalente a 87.000 €.

Un ingeniero hoy con 40.000 € brutos soporta una carga total del 22% (16% IRPF + 6,5% SS) y le quedan apenas 31.000 € netos.

Mi padre vivía con casi tres veces más renta disponible.

En el mismo país. En la misma ciudad.

¿Que hoy se paga menos porcentaje? Lógico, ganando un tercio, claro que el porcentaje baja.

Por el camino, eso sí, se cargaron las deducciones que protegían a la clase media como por ejemplo la deducción por vivienda habitual que desapareció para nuevas compras en 2013.

Y si por algún milagro alcanzas hoy los 120k equivalentes que cobraba mi padre, soportarías un 35% de carga fiscal total en vez de su 27%.

Ocho puntos más por el mismo sueldo real.

¿De verdad vivimos mejor?

Los datos dicen una cosa. La narrativa que nos venden, otra.

[Pau Sabria @kuap]

@swyx one ships with agents. the other thinks it's still 2023 and writes bash scripts.

[Pau Sabria @kuap]

@davidsenra @DavidBaszucki The gap between "exit" and "what's next" is one of the most disorienting stretches of a founder's life. Nobody talks about it. The identity crisis is real.

And apparently optional if you're Baszucki.

[Pau Sabria @kuap]

@TechCrunch Sold a company in 2016.

If I'd sold it in 2026 the entire negotiation would've been done by two AI agents arguing over a term sheet. The pace of value creation in dev tools right now is genuinely unprecedented.

[Pau Sabria @kuap]

@martinvars Again, that position is being lazy.

If residency cannot be managed to adapt to the nuances of the regions, next one could claim the education curriculum should also be simplified and only Spanish taught in the classes.

Slippery slope.

[Pau Sabria @kuap]

@SebJohnsonUK @ProjectEurope_ @Kitty_Mayo_ 100%. Been building here for the last 3 years. The combination of quality of life, growing talent pool, and proximity to European capital is hard to beat. Still early but the momentum is real.