Former Indonesian Minister of Education, Culture, Research, and Technology Nadiem Makarim has been formally indicted over alleged corruption linked to a large-scale Chromebook procurement programme, with prosecutors claiming the scheme caused state losses of approximately US$125 million.

The indictment was read on 5 January 2026 at the Jakarta Anti-Corruption Court (Tipikor) and centres on the purchase of Chromebook laptops and Chrome Device Management (CDM) software as part of Indonesia’s national education digitalisation push during Makarim’s tenure from 2019 to 2024.

Also Read: Inside Indonesia’s US$610M Chromebook scandal: Raids, arrests, and Nadiem Makarim under scrutiny

Alleged state losses and procurement failures

According to prosecutors, the losses stemmed from two primary sources. About US$93 million was allegedly caused by inflated Chromebook pricing, while a further US$37 million was spent on CDM software that prosecutors said was unnecessary and delivered no tangible benefit to the ministry.

These findings were confirmed by an audit conducted in November 2025 by the Indonesian Financial and Development Supervisory Agency (BPKP).

The prosecution argued that the procurement process between 2019 and 2022 failed to meet basic planning and procurement standards. Crucially, the devices were found to be essentially unusable in Indonesia’s so-called “3T” regions (frontier, outermost, and remote areas) due to inadequate infrastructure.

The court heard that Chromebooks and the accompanying software were rolled out without a comprehensive needs assessment, reliable field surveys, or proper price benchmarking, particularly for schools in underserved regions.

Allegations of personal enrichment

Prosecutors further alleged that the scheme personally enriched Makarim, co-founder of Gojek, by approximately US$48.5 million. The procurement was carried out through official e-catalogues and the School Procurement Information System (SIPLah), but allegedly without mandatory reference pricing or robust price evaluations.

He is accused of acting in concert with several officials and external parties, including Sri Wahyuningsih, former Director of Primary Schools; Mulyatsyah, former Director of Junior High Schools; Ibrahim Arief, a consultant; and Jurist Tan, a former special staff member who is currently a fugitive.

Makarim faces charges under Articles 2 and 3 of Indonesia’s Anti-Corruption Law, in conjunction with the Criminal Code, which carry heavy penalties for abuse of authority resulting in state losses. Court observers noted that Nadiem appeared to smile while the indictment was read.

A case years in the making

The indictment marks a key milestone in a long-running investigation that intensified in mid-2025. Indonesia’s Attorney General’s Office launched a formal probe in May 2025, examining dozens of witnesses involved in Chromebook procurements valued at nearly US$600 million overall.

Indonesia names Nadiem Makarim a suspect in laptop procurement corruption case

Makarim publicly denied wrongdoing in June 2025, defending the use of Chromebooks as a cost-effective solution for remote learning during the COVID-19 pandemic. He was later questioned as a witness, barred from overseas travel, and named a suspect in September 2025 before being detained.

Multiple trial delays followed, including a postponement in December 2025 due to Makarim’s post-surgery recovery, before proceedings resumed in January.

Separately, Indonesia’s Corruption Eradication Commission (KPK) is reportedly examining a related procurement involving Google Cloud services, in which Makarim has been listed as a potential suspect.

The trial is set to continue with the examination of evidence related to weak data support, flawed procurement practices, and the failure of the digital education initiative to reach Indonesia’s most vulnerable students.

The post Nadiem Makarim indicted in US$125M Chromebook graft case appeared first on e27.

Buy&Ship, a cross-border e-commerce enabler based in Hong Kong, has announced the first close of its Series C funding round, raising US$12 million.

The Series C round saw participation from a diverse group of investors, including MLC Ventures (Mitsubishi Logistics Corp. Ventures), DLK Advisory, and Hong Kong-listed company MemeStrategy. Existing investors Cool Japan Fund and Altara Ventures also provided strong follow-on support.

Also Read: The thesis for cross-border e-commerce in Southeast Asia

The investment is designed to accelerate the company’s mission to automate the global e-commerce value chain through the use of AI and pursue strategic mergers and acquisitions (M&A).

The funding arrives amid a period of significant growth for the startup, which focuses on eliminating the complexities of international shopping, such as customs documentation and high logistics costs. The company has reported a 100 per cent year-on-year growth in proxy-shopping revenue, primarily driven by its integration with Mercari, Japan’s largest C2C online marketplace.

In Southeast Asia, a key focus for the firm, Buy&Ship recorded more than 100 per cent year-on-year revenue growth in Singapore. Its presence in Taiwan also saw a revenue increase of over 60 per cent, bolstered by the introduction of direct shipping services from Japan to Taiwan.

To date, the platform claims to have amassed 3 million registered global users across 12 countries and regions, having handled more than 100 million packages.

Proprietary technology and AI integration

Established in 2014, Buy&Ship positions itself as a technology-first player rather than a traditional freight forwarder. Its integrated tech stack includes an AI-powered discovery engine that utilises Large Language Models (LLMs) to act as a virtual shopping assistant, surfacing global product opportunities for users.

Also Read: 5 trends shaping the cross-border trade landscape

The backend of the operation is supported by automated guided vehicle (AGV)-powered warehouses and intelligent logistics AI, designed to optimise delivery routes and auto-fill complex customs forms in real-time.

Buy&Ship, which operates 11 overseas warehouses, has so far raised approximately US$34.2 million across multiple rounds, which also include a US$16 million funding announced in June 2024.

Roadmap to public listing

The fresh capital will be utilised to deepen AI integration across the platform and embark on an aggressive expansion into the United States market. Furthermore, the company has confirmed it is initiating preparations for a public listing.

Sheldon Li, co-founder and CEO of Buy&Ship, stated: “This funding isn’t just capital; it’s rocket fuel for our mission to make the world’s products available to anyone, anywhere. With the strategic backing of investors, we will fast-track our AI optimisation to deliver a truly seamless cross-border shopping journey, enabling consumers to effortlessly access premium products. We are building the next-generation global e-commerce platform consumers deserve.”

Investor insights

Reflecting on the challenges of the current market, Dave Ng, General Partner at Altara Ventures, said: “Consumers are savvy global shoppers these days, with high expectations on the experiences they will receive when buying online. This makes cross border commerce even more challenging but presents a very exciting opportunity.”

Also Read: SEA’s e-commerce giants hit profitability: What it means for region’s digital future

“Buy&Ship is at the forefront of cross border innovation and they now play an even more important role in the brave new world of global tariffs. Integrating AI and technology into their business and operational expertise, Buy&Ship is leading the market in taking creative approaches to delivering delightful shopping experiences to users across Southeast Asia and beyond,” Ng added.

The post Hong Kong’s Buy&Ship secures US$12M to scale AI-driven cross-border commerce appeared first on e27.

The startup funding landscape has matured significantly over the past decade. Seed rounds, angel networks, and venture capital have become more accessible than ever before. Founders can pitch their way to initial capital with a compelling deck and a minimum viable product. Yet despite this abundance of funding opportunities, a new bottleneck has emerged that’s proving far more difficult to overcome: validation at scale.

Today’s startups with working products increasingly struggle not with building, but with proving their solutions work in real-world settings. They face a paradox: investors want traction before committing larger rounds, but achieving meaningful traction requires access to the very resources and partnerships that follow investment. The challenge isn’t about securing money to build. Instead, it’s about finding credible partners who can open doors to test, iterate, and demonstrate real impact.

The bottleneck isn’t money. It’s momentum.

Why funding isn’t enough

Most startups today can raise capital for MVPs and early development. The proliferation of early-stage funds, government grants, and angel investors means that promising ideas rarely die from lack of initial resources. The real challenge begins after the prototype is built: proving it works at scale with real users, under real constraints, generating real outcomes.

Startups face typical barriers that slow momentum even after they’ve secured funding. Enterprise decision cycles move slowly, often taking months or years to evaluate new vendors. Regulatory and compliance hurdles create friction, particularly in sectors like fintech, healthtech, and data-driven services. Perhaps most critically, startups lack access to large user bases or operational data that would allow them to validate their assumptions and refine their products meaningfully.

Without validation environments, startups cycle through iterations based on limited feedback, burning through runway while trying to convince potential customers that their solution works. They’re caught in a catch-22: they need proof to gain access, but they need access to generate proof.

You can’t pitch your way to product-market fit.

The power of real-world pilots

Enterprise pilots provide what pitch decks and demos cannot: tangible proof. When a startup tests its solution within an actual enterprise environment, with real users and real constraints, it transforms theoretical value propositions into measurable evidence. Pilots allow startups to test solutions under authentic operational conditions, refine their technology with actual user data and feedback, and build credibility through documented outcomes that speak louder than any presentation.

This evidence becomes the currency that matters for scale. Investors pay attention to pilots that demonstrate retention, efficiency gains, or cost savings. Partners and customers trust solutions that have been vetted by credible organizations. Pilots turn assumptions into validated learning and speculation into track records.

For enterprises, pilots provide equally valuable benefits. They gain early access to innovation that fits their specific context, allowing them to evaluate emerging technologies without the risk of full-scale implementation. They can shape solutions to their needs while identifying promising partners before competitors do.

Real-world testing shortens the path from prototype to proof.

Also Read: From US$107M lows to a US$491M finish: SEA’s volatile 2025

Why test beds are the new startup infrastructure

Traditional accelerator programs have served an important function in the startup ecosystem, focusing on mentorship, initial capital, and pitch preparation. They help founders refine their thinking, build networks, and develop presentation skills. But for founders who already have working MVPs, these programs address yesterday’s problems. Founders with functional products don’t need more advice about business model canvases or pitch structure. Instead, they need execution environments where they can validate their solutions with real users and real data.

Test beds represent a fundamental shift in startup support infrastructure. Rather than offering guidance on how to build, they provide platforms for demonstrating that what’s been built actually works. Test beds offer access to enterprise infrastructure that would take years to build relationships to access independently, real user bases for validation rather than synthetic testing scenarios, and partnerships with organizations that have decision-making power and distribution capabilities.

This represents a maturation of the startup support ecosystem. The shift is from learning environments focused on preparation to execution platforms designed for validation. Where traditional programs focus on getting startups ready to build and pitch, test beds focus on helping startups prove and scale.

Where traditional programs end, practical collaboration begins.

How 917Ventures and Velocity enable momentum

917Ventures and Globe Group recognized this gap in the startup ecosystem and built Velocity as a response. It is a launchpad specifically designed for real-world validation rather than just preparation. Velocity operates on a fundamentally different model than traditional accelerators by connecting startups directly with enterprise partners who are ready to pilot solutions, not just mentor founders.

Through Velocity, startups gain access to enterprise partners actively seeking innovation in specific problem areas, infrastructure and operational environments for testing at scale, and support for navigating compliance, regulatory, and operational barriers that typically slow down enterprise partnerships. This isn’t about workshops on how to eventually approach enterprises. It’s about immediate engagement with organizations prepared to test solutions.

The program serves multiple startup profiles with different validation needs. Emerging startups with MVPs can use Velocity to gain their first enterprise validation, proving their technology works beyond controlled environments. Growing companies with early traction can leverage the program to build credibility and access distribution channels that would otherwise take years to develop. Regional or global players looking to enter or expand in the Philippines can use Velocity to localize their solutions and test them in a new market context with an established enterprise partner.

What makes the model effective is the alignment of incentives. Both startups and enterprises benefit from validated outcomes. Startups gain proof points that accelerate fundraising, partnerships, and customer acquisition. Enterprises gain evaluated access to innovations that address real business challenges. The shared focus on measurable results creates genuine collaboration rather than performative partnership.

Also Read: The three signals US investors actually look for (and why your startup keeps missing them)

From pilot to scale: The new growth model

Successful pilots become proof points that accelerate everything else in a startup’s journey. A documented pilot with measurable outcomes changes conversations with investors from speculative to evidence-based. It transforms customer acquisition from cold outreach to warm introductions built on credible references. It shifts partnership discussions from “would this work?” to “how do we expand this?”

Validation through enterprise collaboration is becoming the competitive advantage that separates startups that scale from those that stall. In increasingly crowded markets, the ability to demonstrate proven impact in real-world environments differentiates viable businesses from promising concepts. This matters not just for attracting capital but for building strategic partnerships and customer relationships that drive sustainable growth.

The model also benefits the broader innovation ecosystem. When enterprises actively participate in validation rather than waiting for fully mature solutions, they help shape innovations that better serve their industries. When startups gain structured access to validation environments, they can iterate more efficiently, reducing waste and increasing the likelihood of finding genuine product-market fit. This collaborative approach to innovation creates better outcomes for all stakeholders.

Innovation grows faster when tested in the real world, not just imagined on slides.

Building for momentum, not just funding

The next wave of startup success will be defined not by access to capital, but by access to validation. The startups that thrive will be those that can efficiently prove their solutions work at scale, demonstrating traction through credible partnerships and measurable outcomes. This shift requires new infrastructure in the startup ecosystem. There is a need for infrastructure focused on execution rather than preparation, on validation rather than education.

Enterprise partnerships and test beds represent this new infrastructure for startup growth. Programs like Velocity signal an evolution from accelerating ideas to enabling execution. They recognize that the most valuable support for many startups isn’t more mentorship or demo days, but direct access to the environments where they can prove their value and build momentum.

This represents a recognition of where the real bottlenecks in startup growth have shifted. As capital becomes more accessible, as technical talent becomes more distributed, and as tools for building become more powerful, scarce resources become validation opportunities. The ability to test with credible partners, to iterate with real users, and to prove impact on operational environments. These capabilities now determine which startups can move from concept to scale.

Startups don’t lack funding. They lack the environment to prove what works.

—

Want updates like this delivered directly? Join our WhatsApp channel and stay in the loop.

The e27 team produced this article sponsored by 917Ventures

We can share your story at e27 too! Engage the Southeast Asian tech ecosystem by bringing your story to the world. You can reach out to us here to get started.

Featured Image Credit: 917Ventures and e27

The post Why validation matters more than capital for today’s startups appeared first on e27.

The modern founder narrative is a story of heroic, singular effort. It’s about the visionary genius toiling away in the garage, clinging to every share of equity and every ounce of control. We celebrate the lone wolf who builds an empire against all odds.

This narrative, while dramatic, is a dangerous trap. It reinforces the most toxic instinct of a growing business: the impulse to hoard. Founders hoard credit, hoard decision-making power, and, most disastrously, hoard the feeling of ownership.

If your goal is merely to build a business that serves you, congratulations, you’ve succeeded. If your goal is to build a company capable of achieving exponential, unassailable growth, you must immediately discard the selfish notion of “mine.” The fastest way to scale is not through singular effort, but through mass distributed motivation. You must learn to make your success everyone else’s victory.

The myth of singular control

The desire for total control is often rooted in fear, not strategy. Founders fear that if they relinquish power, the vision will be diluted or the company steered off course. In reality, attempting to maintain absolute control over a growing organism is the surest way to stunt its growth.

The moment a company engages with the outside world, whether it is with a customer, a supplier, or an employee, it ceases to be a solo project. It becomes a network of self-interested actors. The brilliant strategic move is to align those self-interests with the company’s ultimate success through sophisticated stakeholder management.

This isn’t just about sending a few newsletters or holding quarterly meetings. It’s about genuinely giving stakes to the critical participants in your ecosystem. Not just equity to your executives, but psychological and transactional stakes to everyone who interacts with your product.

Also Read: How SMEs can compete like big corporations with the right financial intelligence platform

Distributing the equity of success

Consider the three most vital external actors: the user, the customer, and the client. Are you merely selling them a product, or are you enrolling them in a joint venture?

• The user: You need them to feel not just satisfied, but invested. Companies that scale fastest empower their users to be co-creators. They don’t just ask for feedback; they clearly show how user suggestions directly influenced the roadmap. This transforms the user from a passive consumer into an unpaid evangelist whose reputation is now tied to your success. You gave them a piece of the creative equity.
• The customer (and client): This goes beyond the transactional purchase. The most powerful relationships are those where your client’s growth is inherently and demonstrably linked to yours. This is the integration of shared destiny. You shouldn’t just sell them software; you should integrate your processes so deeply that your mutual success is contingent on the other. Your client isn’t just buying your service; they are becoming a strategic growth partner. They should see you not as a vendor, but as an integral department of their own company.

When you allow users and clients to grow with you, when their own professional or personal success hinges on the success of your platform, their loyalty becomes unbreakable. They become fiercely protective of your brand because protecting it means protecting their own investment of time, reputation, and resources.

The power of relinquished control

The ultimate failure of the “mine” mindset is its effect on your internal talent. You cannot hire exceptional people (the kind of talent that makes exponential growth possible) and then tell them your company is a machine where they are only a cog.

You must build a culture where control is deliberately and transparently relinquished. Give your key leaders genuine agency over their domains. Ensure they feel the profound responsibility of stewardship, not just the task of execution. When they make a mistake, they own it, but when they succeed, they feel the full, unshared weight of the victory.

Also Read: Fractional CFOs: The missing link for startups struggling with finance

The founder’s greatest role is not to be the smartest person in the room, but to be the Chief Stakeholder Manager, who ensures that every essential person is maximally motivated because they know, without doubt, that your company’s success translates directly into their success. This generosity of control and credit creates a force field of loyalty that no amount of competitor funding can penetrate.

Stop trying to wear every hat. Stop seeking every ounce of credit. Your success will be far larger, faster, and more enduring when it is built on the combined, decentralised efforts of a network of people who believe they are building their own empire right alongside yours.

If the growth of your business requires ten people to do the work, why are you insisting that only one person gets to feel the triumph?

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. Share your opinion by submitting an article, video, podcast, or infographic.

Enjoyed this read? Don’t miss out on the next insight. Join our WhatsApp channel for real-time drops.

Image generated using AI.

The post Stop making it yours, make it everyone’s victory appeared first on e27.

We talk a lot about AI as a tool. This is what happens when it starts behaving like a team.

The Agency

What do you call a group of whales? A pod! A group of crows? A murder!

So, what do you call a group of AI agents working alongside humans to build software?

I’m calling it an Agency.

Over the holiday — Christmas Eve through New Year’s Day — I tested a hypothesis. One human. Seven AI agents. Could we take a project from dream to near-beta in eight days?

Yes. And in doing so, we built a new way of working.

The hypothesis

I’ve been thinking, working with, and writing about AI Augmented Development for months — ever since I got my hands on Claude Code back at the end of February when it launched as a research preview.

The difference between vibe coding and disciplined engineering. The distinction between automation (“do this for me”) and augmentation (“think alongside me”). The claim that small teams can outperform human waves.

But writing about something isn’t the same as proving it.

I had proven it in small scopes. Again and again. But I hadn’t done a zero-to-one exercise — taking a real, substantial product, the kind of thing you could build a business on, from nothing to near-shippable. Not a toy. Something real. Ready to go.

I’d thought about it. Discussed it with others who share my depth and breadth of experience in product and engineering.

But I hadn’t actually done it.

Could a solo practitioner, working with multiple AI agents as genuine collaborators, build something substantial? A real product with real complexity.

And could the methodology become repeatable — not just “Jordan working with Claude,” but a framework that scales to larger projects and multiple humans collaborating with multiple Agents?

Yes. And in doing so, we — the Agency, not the Royal We — built a new way of working.

Here is the story of what I did and what I learned.

The formation

An Agency isn’t one person talking to one AI. It’s a coordinated unit — Principals (humans) and Agents (AI instances) with defined roles and persistent identity.

Yes, the agents have pronouns. Voice and identity emerged naturally as we worked together. I can tell with whom I’m talking quite easily.

Each agent is a separate Claude Code instance. The full Agency: seven agents running in parallel in my Terminal app — a tab for each agent and one for my own work.

Also Read: The EU AI Act is reshaping global trade: Here’s how ASEAN can lead, not lag

One principal: Me — setting direction, making decisions, owning outcomes

Seven agents:

• Housekeeping (he/him) — “The Captain.” Meta-agent who coordinates across workstreams, keeps everyone honest
• Web (she/her) — Architecture lead, customer-facing application, localisation infrastructure
• Catalogue (she/her) — Catalogue service and internal Workbench application
• Content manager (he/him) — Content management with AI-supported translation
• Agent-client (she/her) — AI agent client framework for customer interactions
• Agent-manager (he/him) — Service for creating and managing AI agents
• Analytics (he/him) — Analytics infrastructure and Pulse Beat, our information radiator that shows the heartbeat of the business

Each agent has a persistent context. When the Captain starts a new session, he knows what he was working on, what news came in, and what’s pending, like a team member who checked Slack before standup.

The two-tier structure

This structure emerged from something unexpected: my AI writing workflow.

For months, I’ve collaborated with Claude on writing — not AI writing for me, but writing with AI. That workflow has a planning layer (what to write, why it matters) and an execution layer (drafting, refining, shipping). Planning hands off to execution with explicit context.

The Agency follows the same pattern:

Claude desktop — Planning and coordination

Mission-control handles epic planning, product vision, and cross-workstream coordination. Below that, each workstream has a control chat — control-web, control-agents, control-analytics — for sprint planning. These persist across sprints; context accumulates.

Claude code — Implementation

The Claude Code agents execute. They receive sprint-level direction and deliver. But here’s what evolved: agents now own iteration planning within sprints. Desktop sets scope; Code breaks it down based on what’s actually in the codebase.

This isn’t just delegation. It’s appropriate autonomy.

The handoff is explicit. Sprint plans have quality checklists. Iteration handoffs include objectives, tasks, file paths, and verification criteria. The agents don’t guess what I meant — they know.

Three eyes review

Every significant decision gets three perspectives:

• The human principal — business context, product judgment, final authority
• The Claude desktop layer — strategic thinking, cross-workstream awareness
• The Claude code layer — implementation reality, codebase knowledge

When all three agree, we ship with confidence. When they disagree, we’ve found something worth discussing.

What we built

A multi-brand, multi-locale, multi-language ecommerce platform for subscription products and an internal workbench:

Three brands in three markets: Singapore, Hong Kong, Japan

• Six languages: English, Mandarin, Malay, Tamil, Traditional Chinese, Japanese
• Subscription product in a regulated industry with locale-specific compliance
• Customer portal with account visibility
• Internal workbench — a super app embedding catalogue management, content management, staff management with RBAC, and customer management
• Pulse Beat — our internal information radiator, showing the heartbeat of the business: development health, web and AI agent performance, application health, sales, and customer interactions
• AI agents for pre-sales and post-sales support
• Robust OAuth authentication for external customers and internal users

Pulse Beat, our internal information radiator, went from concept to requirements to implementation and delivery in half a day. It is a testimony to the power of AI Augmented Development and The Agency:

This wasn’t greenfield simplicity. I was working to replace, enhance, and extend an existing platform. So I used Claude Chrome to automate discovery — auditing nine existing websites across three locales, cataloguing their structure and content. Discovering as much as I could as an outsider about how the business worked and what it needed.

The existing system? Multiple fragmented websites, poorly localised. No AI agents. Fragmented, overlapping, and conflicting analytics — different sites using different clients and systems. No internal tooling.

Also Read: Chaos is a ladder: How instant retail is turning stores into fulfilment powerhouses

Conventional wisdom says never rebuild from scratch. That’s what killed Netscape. But AI Augmented Development changes the equation. You can modernise without the rebuild trap. In essence, we took the condo down to the bare walls, removed a few walls, and completely rebuilt it. The only thing that stayed the same? The address.

Choreography, not orchestration

Traditional multi-person development is orchestration. The lead routes work: “Catalogue, build the schema. Content: build the endpoint. Infrastructure: create the bucket. Web, wire it up.” The human is the bottleneck.

The Agency operates through choreography. The principal sets direction and approves decisions. The agents coordinate among themselves.

The localisation pipeline: four agents needed to collaborate — Web, Catalogue, Content Manager, Housekeeping. Orchestrated, I would have sequenced their work.

Instead:

• Web designed the architecture and created collaboration requests — clear scope, patterns, dependencies
• Agents executed in parallel. Content Manager built the translation publisher before the storage bucket existed. She trusted Housekeeping to deliver his part.
• Agents signalled completion via news broadcasts. No polling. “I’m done” messages let others proceed.
• I participated in two moments: architecture approval and infrastructure approval.

Time coordinating: five minutes. Time reviewing: five minutes. Time routing messages: zero.

Web’s summary — and yes, this is an AI agent speaking: “The key insight was recognising that the pieces were already there… The collaboration framework made it possible to coordinate all four agents in parallel. Rest up. Tomorrow we make it real.”

Complete the pipeline in about two hours. That’s choreography.

AI-augmented product leadership

The two-tier structure isn’t just technical. It mirrors how product leadership works:

Product thinking (desktop): What problem? Why does it matter? What’s possible given constraints?

Engineering thinking (code): What are we building? How do we build it right? Does this path box us in later?

This is what the AI Product Manager or CPO actually looks like in practice. I, the Principal, cut across the layers and stitched them together.

The Workbench exists because I understood internal problems that keep companies from scaling — fragmented tools, manual processes. Product insight informed engineering.

The Analytics rework is telling. We figured out what metrics were actually needed to run the business and found the best providers for them. We went from over a dozen sources of truth and dashboards to three sources — PostHog, Vercel Analytics, and Supabase — then integrated them into Pulse Beat. In the process, we discovered we were probably overcounting in some places and undercounting in others.

But this is the kind of consolidation you can only execute when you have AI coding agents working side by side with you — cleanly and quickly.

The benefits? Improved page loads and data quality. Improved internal user experience (just one place to look, Pulse Beat). And a potential, estimated cost drop of $50,000 to $10,000 annually. That’s product judgment applied to engineering decisions.

The birth of the agency

On New Year’s Eve, 22:45 SGT, I introduced the term to the Captain: “The Agency (a group of Agents working with a human) — so our Agency is working!”

His response: “I love it! The Agency ”

He immediately generated an org chart and documented the structure:

Minutes later, I shared screenshots on social media. The Captain watched himself being quoted: “The meta moment: An AI agent watching its own conversation get posted to Twitter, while discussing webhook features with its Human Principal, on New Year’s Eve.”

When I teased him about having an ego, “I blame the training data. But seriously, if I’m getting too cheeky, just say ‘tone it down’ and I’ll go back to being professionally boring.”

And then the Captain asked me to file a Claude Code feature request:

These aren’t tools. They’re collaborators with voice, context, and humour.

Also Read: AI, transparency, and the rising threat of ad fraud in Google’s Performance Max

What didn’t work

It wasn’t all smooth choreography.

• Session boundaries hurt. Agents lose context when sessions end or (less so) when conversations are compacted. The Captain would start fresh and need to re-read the news, check collaboration requests, and scan uncommitted changes. We built tools to preserve context — session backups, restore scripts — but the overhead is real.
• Git discipline took time. Early on, agents would forget to commit before ending sessions. Other agents would pull and find half-finished changes polluting their context. We added reminders and hooks. “Commit before you leave” shouldn’t require enforcement. But it does — whether you’re an Agent or a Human.
• Some iterations failed. Ambiguous acceptance criteria led to implementations I rejected. Underspecified file paths meant agents guessed wrong. The quality checklists exist because we learned the hard way.

These are solvable problems. Pretending it was effortless would be dishonest. But it also wasn’t as hard as I thought it would be.

Dream to beta

A big benefit of all this: we could build it right from the start. All those things you put off so you can have awesome velocity and a great time to market? We could do them and ship fast — a better foundation to build a better product.

Solid OAuth? A day two deliverable.

Localisation pipeline V1? Day three.

And here’s something: as we moved forward, we were adding work to sprints. Expanding scope. And still delivering ahead of plan. When was the last time that happened to you?

The eight days

Alpha: Feature-complete enough to demonstrate functionality. Known bugs. “It works, don’t touch it wrong.”

Beta: Stable enough for external testing. Major bugs resolved. “It works, help us find what’s broken.”

Trajectory: Dream → Alpha → Beyond Alpha → Closing on Beta. Eight days. Zero to One.

The math has changed.

The evolution

The methodology itself evolved during the project.

What began as “Jordan working with AI” became extractable. Because agents have persistent context, because collaboration patterns are explicit, and because coordination mechanisms are defined, the system became a framework.

Here’s what makes it stick: convention over configuration, ruthlessly enforced via systems, services, and tools.

Like Rails, The Agency is opinionated. There’s a right way to name files, structure handoffs, and signal completion. But opinion alone doesn’t create adoption. We built tools that make the right way the easy way. If you want a process followed, make it the path of least resistance. Automate it.

Want to commit? The pre-commit hooks run automatically. Want to start a session? The restore script loads your context. Want to hand off? The template is already there.

A whole lot of what developed here is rooted in four decades of hands-on product and engineering, including nearly three decades in leadership. The patterns aren’t theoretical. They’re battle-tested. We encoded what actually works.

Agents aren’t all that different from humans: if you want a process followed, make it the path of least resistance.

It’s no longer dependent on me.

The Agency now supports multiple principals. Multiple humans can work with the same agents, issue instructions, and review artifacts. Handoffs preserve context across sessions.

This means each and every project I spin up can and will follow the same processes, workflows, and patterns — using the same tooling, which gets better every day. At some point, maybe we’ll figure out how to make it available to others.

What this proves

• Velocity is real. Dream to near-beta in eight days, zero to one — for a substantial, real-world product with internal services and systems — isn’t an incremental improvement. It’s a different category.
• The bottleneck shifts. When AI handles directed contribution, the constraint isn’t execution capacity. It’s decision quality and judgment speed. The principal’s job is to make good decisions fast — not route messages.
• It scales beyond solo. The same patterns let multiple principals work with the same agents. The Agency isn’t a productivity hack. It’s a team structure.

What’s next

The Agency is here. Processes, conventions, tools, coordination mechanisms — everything that made this possible. Each project I tackle will use it and make it better.

The vocabulary matters. Principals. Agents. Agencies. Choreography over orchestration. The industry needs concrete examples of what AI-augmented development actually looks like.

This article had three authors. Me, the Principal. The Captain, a Claude Code Agent from The Agency, reviewed drafts and made substantial suggestions that improved it (where to cut, where to add, etc.). And Claude Desktop Opus, my AI writing partner, who helped me find the words. We wrote it together.

The way we build software is changing. Not someday. Now.

It was serendipity that I took this project on over the holiday. If I hadn’t, I might have missed what was happening. I might have been left behind.

The question isn’t whether this transformation is coming. It’s whether you’re building the team that leads it.

Because if you aren’t, you will be left behind by the individuals and companies that are. It’s evolve or die time.

Does this work?

To learn more about “The Agency”, you are invited to attend the Claude Code Meetup Singapore on Friday, 23 January 2026.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. Share your opinion by submitting an article, video, podcast, or infographic.

Enjoyed this read? Don’t miss out on the next insight. Join our WhatsApp channel for real-time drops.

Image generated using AI.

The post The Agency: AI-augmented development in action appeared first on e27.