Insights

Notes on building AI-native companies.

What we have learned building products, teams and go-to-market engines with people and agents working together.

Founders

A company is a hundred jobs.

Connectify · 4 min read

Write down everything a young company has to do in its first year. Customer interviews. A market map. Pricing. A product spec. Design. Code. Tests. Security. A launch plan. Content. Sales outreach. Finance. Hiring. Support. The list passes a hundred before lunch.

Most founders start with one or two of those skills and a great deal of conviction. Conviction is necessary, but it does not write the test suite or the investor update. So the founder either hires before there is money, outsources piece by piece to people who never meet each other, or tries to do it all and does most of it badly.

Good ideas rarely die because they are wrong. They die because one person cannot do a hundred jobs.

Two things have changed. First, the cost of much of that work has collapsed. AI can now research a market, draft a spec, write and test code, and prepare outreach at a fraction of last decade's cost. Second, it can do that work continuously, without waiting for a hire.

But cheaper work is not the same as good work. Someone still has to decide what matters, check the output, and own the result. That is why we build teams, not tools: a lead for each function, accountable for decisions, with agents doing the volume beneath them.

The practical result for a founder is simple. On day one you get a product lead, an engineering lead, design, strategy, marketing and sales, each multiplied by agents. You spend your time on the decisions only you can make.

↑ Back to all insights
Method

Go or kill: the eleven gates.

Connectify · 5 min read

We take every company through eleven stages, from the first idea to the first billion: ideate, brainstorm, finalise, design, build the MVP, launch, product–market fit, the go-to-market engine, scale, blitzscale and the first billion.

Each stage ends at a gate. A gate is a short written document and a decision: go, change course, or stop. The structure borrows from methods used by top strategy firms and venture builders: issue trees to break a problem down, hypotheses with explicit kill conditions, and stage-gated investment so money follows evidence.

Three rules make the gates work:

  1. Criteria first. The bar for passing is agreed before the work starts, so nobody moves the goalposts after seeing the results.
  2. Evidence in writing. Every claim in the gate document links to its source: interviews, data, tests. If it isn't written, it isn't decided.
  3. A person signs. Agents prepare the memo. A named lead and the founder make the call.
Killing an idea at gate three is a success. It saves the years and money that gate eight would have cost.

Founders sometimes worry that gates slow things down. In practice the opposite happens. When everyone knows what "done" looks like for a stage, work moves faster, arguments are shorter, and investors can see exactly where the company stands.

↑ Back to all insights
AI & teams

Agents do the volume. People make the calls.

Connectify · 4 min read

Our thirteen agents each have one job. SCOUT maps markets. ECHO turns interviews into jobs-to-be-done. FORGE writes code and tests. SENTRY tries to break every release. Each one reports to a person.

The split is deliberate. Agents are very good at volume: reading two hundred reviews, drafting forty personal messages, running four hundred tests overnight. They are not accountable for consequences, and they should not pretend to be.

So we draw the line in the same place every time:

  • Agents research, draft, build, test and report.
  • People decide anything that affects a customer, money, a hire or a live system.
  • Nothing ships, publishes, spends or sends until a named lead signs it off.
Speed without accountability is just risk, delivered faster.

This is also what makes AI work trustworthy for enterprises and governments. Every agent action is logged, every decision has a name next to it, and the audit trail exists from day one rather than being reconstructed later.

See what each agent does and the rules it works under →

↑ Back to all insights
Government

Building AI for government: sovereign by default.

Connectify · 5 min read

Citizens now judge every service against the best app on their phone. Government teams know this, and most want to move faster. What holds them back is rarely ambition. It is the fair question: is it safe?

Our answer is to make the safe choice the default, not an option:

  • Data stays in India, hosted where the department chooses: a state data centre, government cloud or on-premise.
  • The DPDP Act is designed in from the first sprint: consent, purpose limitation and deletion.
  • Every action is logged for audit, and incidents follow the CERT-In reporting window.
  • Officers decide. Agents prepare files and route requests; no automated decision touches a citizen without a person.
  • No lock-in. Source code and documentation are handed over.

Language and access matter as much as security. A service that works only in English, or only on a new phone, is not a public service. We build for Kannada, Hindi, Marathi and more, to GIGW and WCAG accessibility guidelines, and for low-end devices on patchy networks.

Start with one district, one ward or one line. Prove value in weeks. Then scale what works.

How we work with government teams →

↑ Back to all insights