In this episode, host Krish Palaniappan speaks with Giacomo Balli, an independent technology advisor who has released more than 100 applications throughout his career. Their conversation examines how artificial intelligence is changing the economics of building software, challenging traditional consulting models, reshaping workforce requirements, and creating new opportunities for businesses. Drawing on their experiences in software development and entrepreneurship, they explore practical questions about organizational structure, technical expertise, AI spending, and the emergence of software ecosystems designed for autonomous agents.
Podcast
Exploring the changing economics of software engineering, the future of technology consulting, AI-driven workflows, and agent-to-agent integrations — on Apple and Spotify.
Building Software Is Easier. Building Reliable Products Is Another Matter.
Giacomo began developing iPhone applications as a teenager in Florence, Italy, eventually turning his interest into a career spanning more than a decade. Today, he helps startup founders and business owners navigate technology decisions, evaluate development proposals, and avoid architectural mistakes. He describes a client who initially had a family member build an application using AI-assisted coding tools. Although the application looked impressive, the client returned several months later seeking professional help because the underlying implementation wasn’t sufficiently reliable for business operations. The experience illustrates the distinction between generating functional code and engineering software that can support a growing business. Technical decisions have consequences extending beyond the initial implementation, including maintainability, scalability, security, and operational reliability. Giacomo’s advisory role focuses on anticipating these consequences and translating between business objectives and technical realities, often helping organizations make important decisions without requiring a permanent executive or engineering hire.
Is Traditional Technology Consulting Becoming Obsolete?
One of the conversation’s central questions is whether organizations will continue paying consultants for work they can increasingly accomplish using the same AI tools. Historically, consulting firms provided specialized expertise, development capacity, project management, and access to talent that businesses couldn’t justify employing full-time. But when software implementation, documentation, research, and analysis become more accessible, consultants must reconsider their value proposition. Krish questions why a company should pay an external consultant to use tools that its own employees could operate directly. Giacomo suggests that consulting will increasingly move toward strategic advisory work, helping organizations determine what to build, evaluate technical decisions, implement effective AI workflows, and organize internal knowledge. Rather than performing every task for a client, consultants may help clients develop the capabilities to execute independently. This creates opportunities for professionals who can combine technical experience with business judgment, while placing pressure on consulting models built primarily around billable implementation hours.
Smaller Teams, Accountability, and the Future of Employment
The discussion examines the implications of reducing organizational layers, particularly middle-management positions and traditional software development roles. Giacomo argues that smaller teams can accomplish considerably more when employees possess strong domain expertise and understand their organization’s priorities. He emphasizes accountability over elaborate management structures, suggesting that companies may increasingly favor individuals who can make decisions, supervise automated processes, and take ownership of outcomes. The conversation also explores whether white-collar professions face greater exposure to automation than jobs requiring physical presence. Software development, technical writing, administrative work, and other computer-based activities are particularly suitable for AI-assisted workflows, while hospitality and similar industries still require people to perform physical services.
Krish introduces an additional economic consideration: industries can experience disruption even when their own operations are difficult to automate. If a consulting engagement previously required ten employees to travel regularly but now requires only two, hotels, airlines, restaurants, and other businesses serving those travelers could experience reduced demand. Giacomo distinguishes this indirect impact from the automation of hotel operations themselves, where housekeeping and guest services still require physical labor. Their exchange highlights the broader economic consequences of changing workforce requirements and raises questions about how employment patterns, business travel, and service industries might evolve.
From Manual Coding to Automated Development Pipelines
Giacomo shares a particularly striking example from his own development work. Although he has largely stopped writing code manually, he reports releasing approximately 30 iOS applications and 40 Android applications over five months using AI-assisted workflows. For certain applications, work that previously required six to eight months can now be completed in a morning. His approach extends well beyond code generation. He describes an automated pipeline covering idea discovery, market research, competitor analysis, user-review mining, feature prioritization, implementation planning, coding, and validation. Multiple agents and models review one another’s plans and outputs, allowing him to supervise the process rather than manually execute every step.
An important part of this approach is capturing knowledge that would otherwise remain undocumented. Giacomo calls this creating a map of work: recording business rules, procedures, technical requirements, and accumulated experience so that agents can operate with organizational context. He also describes changing his approach to repetitive tasks. Previously, he might automate something after performing it several times; now, he aims to capture the process immediately so it can be reused. The resulting system moves human involvement toward defined checkpoints for review, verification, and decision-making. His experience illustrates how development productivity can depend as much on workflow design and knowledge management as on the capabilities of individual models.
The Economics of AI: Employees, Tokens, and Return on Investment
A substantial portion of the conversation focuses on a difficult financial question: how should companies balance spending on human employees against spending on AI tools and tokens? Krish considers hypothetical startups choosing between hiring additional developers and allocating larger budgets to AI infrastructure. He also connects the question to the enormous capital expenditures of major technology companies, whose investments in AI infrastructure have raised investor concerns about profitability and future returns. For smaller organizations, the challenge is more immediate: determining whether additional model usage creates enough value to justify its cost.
Giacomo acknowledges that there is no straightforward formula for measuring AI’s return on investment. Productivity improvements may be visible, but translating them into financial outcomes depends on the business, the task, and the organization’s growth strategy. A company might use AI to reduce staffing requirements, accelerate its existing roadmap, or pursue additional projects without increasing headcount. Each approach produces a different economic result. He also cautions against optimizing token costs too early in the adoption process. Organizations still experimenting with AI should first identify workflows that create meaningful value, then improve efficiency through techniques such as selecting less expensive models or adjusting reasoning requirements. His concern is that aggressively minimizing initial expenditure could prevent companies from discovering opportunities whose benefits substantially exceed the cost of experimentation.
Agent-to-Agent APIs: A New Model for Software Integration
The conversation turns to the changing role of APIs as AI agents become more involved in software development. Krish explains how Snowpal’s API-first business, which offers several RESTful APIs through AWS Marketplace, is adapting to this shift. Traditionally, developers read documentation, selected endpoints, obtained API credentials, and wrote integration code. Increasingly, users can describe a business problem to an AI agent and rely on it to identify suitable APIs, generate the necessary code, and combine services from multiple providers. The developer may not need to understand every underlying endpoint or even know which provider supplies a particular capability.
This creates new requirements for API businesses. Documentation must be accessible not only to human developers but also to agents that need to discover services, understand their capabilities, and integrate them correctly. Krish discusses the importance of adapting API descriptions, documentation formats, and supporting files for agent consumption. Giacomo shares related work converting projects into MCP-compatible tools and plugins, reflecting growing interest in agent-accessible software interfaces. Both recognize that agent-to-agent interaction could change how software services are discovered, distributed, and consumed. For API providers, making products understandable and usable by autonomous systems may become an important part of their distribution strategy.
What Happens to Technical Expertise?
The conversation ultimately raises a broader question about the skills developers, consultants, and technology professionals should cultivate. Krish describes his own movement away from manually writing code and toward reviewing generated implementations, evaluating design patterns, identifying potential problems, and supervising agents. Giacomo’s experience reinforces the importance of technical judgment, domain expertise, and the ability to organize complex workflows. Professionals who understand business requirements and can evaluate the quality of AI-generated work may be better positioned than those whose value depends primarily on executing repetitive technical tasks.
The discussion also acknowledges uncertainty. Neither participant claims to know precisely how consulting demand, employment levels, or organizational structures will evolve. What emerges instead is a practical framework for adapting: understand the business problem, develop strong technical judgment, experiment with automated workflows, capture institutional knowledge, and evaluate investments against measurable outcomes.
Conclusion
The conversation highlights several interconnected changes in the technology industry: software implementation is becoming more accessible, traditional consulting models face pricing pressure, organizational structures may become leaner, and APIs are evolving to serve autonomous agents. Yet easier implementation does not eliminate the need for reliable architecture, sound business decisions, or accountability. Giacomo’s development experience demonstrates the potential of coordinated AI workflows, while the discussion of consulting economics and token spending reveals the challenges businesses must still resolve.
For founders, developers, consultants, and technology leaders, the central question is no longer simply how quickly software can be produced. It is how to build dependable products, organize work efficiently, allocate resources intelligently, and develop expertise that remains valuable as implementation costs decline. The competitive advantage may increasingly belong to those who understand what needs to be built, why it matters, and how to deliver it reliably.